An investigation into attentional blink the attentional engagement hypothesis

... attenuation of the AB, suggesting that understanding the effects of the blank on the blink is crucial to an understanding of the underlying cause of the temporal limits of attention In this thesis,... similarity between T1 and the lag distractor The results from Experiment support the attentional engagement hypothesis In all these experiments, the lag distractor was systematically manipulated All... all lags, with the exception of lags and 5.26 An analogous ANOVA revealed that the T2 performance of the repeat-T1 conditions was also much better than the baseline condition, and F (1, 14) =

An Investigation Into Attentional Blink -- The Attentional Engagement Hypothesis Tan Wah Pheow (B. Soc. Sci. (Hons.), NUS) A THESIS SUBMITTED FOR THE DEGREE OF MASTER OF SOCIAL SCIENCE DEPARTMENT OF PSYCHOLOGY NATIONAL UNIVERSITY OF SINGAPORE 2005 Acknowledgements This thesis would have been impossible without the following people: Dr Chua Fook Kee, my supervisor. Because without your patience and guidance, this thesis would never have been completed. My parents, Mr. Tan Tau Tin and Mrs. Tan Chu Moi, who had supported me wholeheartedly in my chosen path in life. My siblings, Tan Wah Hao and Tan Qiuting, for bearing with your crazy elder brother. All my friends who had to bear with my eccentricity and madness, and being there for me when I was down, including: Chai Chengkuo, whom attempted to test the human body’s endurance limits to alcohol with me. Goh Swee Guan and Khoo Seow Feng, for all those Friday nights where we talked about our lives over supper and mahjong. Wee Tze Yuan, for all those sleepless nights where philosophy, science, love and life dominated our conversations. Sharen Sim and Ho Lifen, for lending a hearing ear whenever I feel down. The guys from HCJC judo club (1995-1997), and all you guys and gals from the psychology honours class (2002/03), may we all succeed in our endeavors in life. The crazy Sunday soccer gang at Block 180 Ang Mo Kio. Thanks for giving out and taking in all those bruising tackles, just to prove that we all come from the school of hard knocks. The guys at heteropoetry club, Fiona Teo Suling, Fang Weicheng, Huang Guangqing, Zeng Zhaocheng, Chiu Weili, Sam Sheen Mun Kong, Apple Hong and Lin Rongchan for all those discussions on poetry and literature, without which my life would be empty. And all those girls who broke my heart and taught me the lessons of love. You will always have a place in my heart. Last but not least, I thank the Singapore Millennium Foundation (SMF) for awarding me the scholarship for me to pursue my Masters. Without their sponsorship, this thesis would not have been possible. I Table of Contents Page Acknowledgements I Table of Contents II Summary III List of Figures IV Chapter 1 1 Chapter 2 21 Chapter 3 40 Chapter 4 79 Bibliography 99 II Summary When participants are required to identify two targets presented in a rapid serial visual presentation (RSVP), identification of the second target is affected when it appears within 500 ms of the first target. This phenomenon has been termed the attentional blink (AB). In the current thesis, the lag 1 distractor is varied in order to manipulate the pattern of AB attenuation. In Experiment 1a and 1b, a repeat-T1 distractor that was identical to T1 was inserted in lag 1. The repeat-T1 distractor was in target and distractor luminance in Experiment 1a and 1b respectively. It was found that inserting a repeat-T1 in target luminance led to an improved T2 performance at lag 2, while this was not found when the repeat-T1 was in distractor luminance. The extant AB models could not account for the pattern of results obtained. A new AB model based on temporal attentional shift (Chua, 2005; Wee & Chua, 2004), the temporal coding hypothesis (Dixon & Di Lollo, 1994), and the theoretical ideas of Loftus and his associates (e.g., Busey & Loftus, 1994) is introduced. This model, named the Attentional Engagement Hypothesis, could account for the data in Experiment 1. The main hypothesis of this model is that AB occurs because attention fails to disengage from a previous target rapidly enough. It is hypothesized that attentional disengagement from a target is modulated by how rapidly the visual system can detect the target’s termination. The argument in this thesis is that target termination is signaled to the visual system when (a) an object change is detected, or (b) the visual system senses that there is no more information available for acquisition from the target. In order to test this new model, a double-stream RSVP presentation was employed in Experiment 2a, 2b and 2c. The lag 1 distractor varied was a repeatT1, a chimeral distractor, and a four-dot distractor for Experiment 2a, 2b and 2c respectively. The findings from these experiments support the Attentional Engagement Hypothesis. There are several implications from the findings in this thesis: (a) it argues for the dissociation between attentional control and stimulus processing; (b) it places the AB phenomenon as an early selection issue; and (c) it argues for a lower boundary of temporal limit for visual attention. III List of Figures Figure Page Title 1 4 Data from Pilot Study Depicting Signature AB Function 2 26 Baseline and Blank Conditions of Single RSVP Stream 3 27 Time Course of Stimulus Presentation 4 27 RSVP Presentation of Repeat-T1 Condition 5 31 T1 Performance for Experiments 1a and 1b 6 33 T2 Performance for Experiments 1a and 1b 7 45 Schematic Correlations of an RSVP Stream 8 55 Schematic Correlations for Single and Double RSVP Streams 9 56 Baseline and Blank Conditions of Double RSVP Stream 10 57 Repeat-T1, Four-Dot and Chimeral Conditions in Experiment 2 11 60 T1 and T2 Performance for Experiment 2a 12 62 Chimeral Distractor 13 64 T1 and T2 Performance for Experiment 2b 14 67 Data From Pilot Study Depicting T2 Performance at Lag 1 IV Figure Page Title 15 70 Four-Dot Distractor 16 73 T1 and T2 Performance for Experiment 2c V Chapter 1 General Introduction In the past two decades, there has been a quickening of research on the temporal characteristics of attention, particularly the distribution of attention over time. According to Shapiro (2001), the underlying time-course of attention provides knowledge of “the temporal availability of whatever property (or properties) of the brain that is (or are) responsible for enhancing perception” (p. 1). 1 The experimental paradigm often used to investigate the temporal characteristics of attention is the Rapid Serial Visual Presentation (RSVP) paradigm. The typical RSVP paradigm requires participants to view a stream of visual items (approximately 10 items per second) all presented in the same location. The targets are embedded within this stream of items. They are demarcated from the rest of the items (i.e., the distractors) in the stream by either physical attributes (e.g. luminance difference) or semantic attributes (e.g. letters amongst digits).1 Participants are required to identify and report the targets at their leisure (but see Jolicœur, 1998). Shapiro (2001) reported that attention is needed to conjoin target-defining attributes (e.g., color) and to-be-reported feature (e.g., the letter’s identity) of a target in a single-task RSVP experiment. This implicitly assumes that attention should be available after target identification, which takes approximately 100 ms (e.g., Lawrence, 1971). However, data from experiments in which observers had to identify two targets (dual-task RSVP) reject this assumption. In a dual-task RSVP experiment, researchers can track the time-course of events following the selection of the first target (Shapiro 2001). Although participants could identify the first target accurately, identification of a second target that appeared within 200 ms to 500 ms of the first target is generally impaired. This identification deficit has been called the “attentional blink” (Raymond, Shapiro & Arnell, 1992). 1 Chun and Potter (1995) argues that using physical attributes to demarcate targets results in the independence between target defining attributes and target features to be reported. Hence, attention is required for the conjunction both sets for features for reporting (Treisman & Gelade, 1980). Chun and Potter argued that it is plausible AB might be due to a conjunction failure rather than processing limitations. Hence, semantic attributes are employed to demarcate targets in order to rule out this account. In their study, Chun and Potter demonstrate that targets demarcated by semantic attributes also results in an AB effect. 2 It is important to note at the outset that the AB effect is an attentional rather than a sensory effect (Raymond et al., 1992). Raymond et al. conducted a control condition in which participants were told to ignore the first target and only to report the second target.2 Here, the identification of the second target was not impaired, implying that the AB effect was not caused by sensory factors, such as low-level visual transients produced by the first target. The failure to identify the second target probably stemmed from attentional processes associated with the identification of the first target. The absence of an AB effect for this control condition has been well replicated (e.g., Shapiro, Raymond & Arnell, 1994; Raymond, Shapiro & Arnell, 1995). In a typical AB experiment, the first and second targets are denoted as T1 and T2 respectively, while the primary dependent variable is accuracy. However, there have also been studies which employed reaction time as dependent variable (e.g., Jolicœur, 1998). The term “lag” refers to the number of items appearing after the first target. The “lag 1 distractor” is the letter appearing immediately after the first target. A second target in lag 4 means that there are three distractors intervening between the two targets. In this thesis, I shall employ these terms when describing experimental procedures. The degree of impairment of T2 identification has often been used as an index of the magnitude of the AB effect. However, the accuracy of T2 identification is generally not taken as the dependent variable. Rather, the accuracy rate of the second target’s identification conditionalized on the first target’s identification (i.e., P[second 2 In Raymond et al.’s (1992) original experiment, the first target was named “target” while the second target was named “probe”. The first target was a white letter amongst black distractor letters, while the second target was always a black “X”. 3 target | first target]) is employed. When the first target is not identified, there is no way of ascertaining whether participants had attended to T1. What is of interest is in T2 identification performance only when attention had been allocated to the T1. The signature AB function is depicted in Figure 1. The data were obtained from a pilot study (N = 10) (Tan, unpublished data). When P(T2|T1) performance is plotted against lag, a U-shaped curve was found. T2 identification performance was high at lag 1 (i.e., known as lag 1 “sparing” effect, [Potter, Chun, Banks & Muckenhoupt, 1998]), but decreases thereafter until it reaches a minimum at lag 2, and then increases steadily until lag 7 where the function asymptotes. This finding has been widely replicated, with different stimulus types (e.g., digits, symbols, words), different stimuli presentation parameters (e.g., different SOA and inter-stimulus interval), and different experimental procedure (e.g., presentation of T1 and T2 masked without intervening distractors) (e.g., Ward & Duncan, 1996; Ward, Duncan & Shapiro, 1997). Figure 1. Data from Pilot Study Depicting Signature AB Function 4 Blank Inserted In Lag 1 The type of distractor inserted in lag 1 modulates the magnitude of the AB effect (e.g., Shapiro et al., 1994; Chun & Potter, 1995). When the lag 1 distractor is more similar to T1, a larger AB effect is obtained (e.g., Chun & Potter, 1995; Raymond et al., 1995). Even when the lag 1 distractor is highly dissimilar to T1 (e.g., dots, blank rectangle, keyboard symbols), the AB effect is merely attenuated, and not eliminated completely (e.g., Chun & Potter, 1995; Grandison, Ghirerdelli & Egeth, 1997; Raymond et al., 1995). The only exception appears to be the situation when a blank is inserted in lag 1. This is a critical finding. This thesis explores this issue. In other AB experiments, it was found that inserting a blank in lag 1 either attenuated AB reliably and drastically (e.g. Chun & Potter, 1995; Grandison et al., 1997), or eliminated it completely (e.g. Raymond et al., 1992).3 Compared to the other types of lag 1 distractor, a blank inserted in lag 1 always resulted in the greatest attenuation in the AB for the given set of conditions within the particular experiment. The blank attenuates the AB only when it is inserted in lag 1 (e.g., Chun & Potter, 1995; Raymond et al., 1992). But if one or more distractors intervened between T1 and the blank, AB is not attenuated even when the blank duration lasted 270 ms (Raymond et al., 1992). This suggests that the crucial factor for both the elicitation and modulation of the AB is the item trailing T1. This postulation is further supported by studies using the “skeletal” RSVP paradigm (e.g., Ward et al., 1996, 3 Chua (2005) offers an explanation as to why a blank at lag 1 sometimes attenuates AB, while at other times eliminates it. However, whether a blank inserted at lag 1 eliminates or attenuates AB is not of central interest in this current thesis. Therefore, this issue is not pursued further here. Interested readers can refer to Chua’s paper for an explanation. 5 1997). The skeletal RSVP paradigm essentially involves presenting T1 at one of two possible locations after which it is masked by a pattern stimulus. T2 is then presented, also at one of two possible locations, and then masked. The critical manipulation was the SOA between T1 and T2. An AB effect was also found for this “skeletal” RSVP paradigm, again demonstrating that even when the item trailing T1 (its mask) was not trailed by any items, the AB was obtained. What is critical appears to be the item trailing T1. In the following sections, I introduce several AB models. I focus on how each model explains why a blank inserted in lag 1 attenuates the AB. McLaughlin, Shore and Klein (2001) classified the various AB models into two broad classes: (a) the interference model (e.g., Raymond et al., 1995; Shapiro et al., 1994); and (b) the bottleneck model (e.g., Chun & Potter, 1995; Giesbrecht & Di Lollo, 1998; Seiffert & Di Lollo, 1997; Duncan, Ward & Shapiro, 1994; Ward et al., 1996, 1997). In this thesis, I adapted McLaughlin et al.’s classification, with the exception that I further separate the bottleneck model into the processing model (e.g., Chun & Potter, 1995; Giesbrecht & Di Lollo, 1998; Seiffert & Di Lollo, 1997) and the attentional dwell model (e.g., Duncan et al., 1994; Ward et al., 1996, 1997).4 4 The reason why I classify the extant AB models into three broad models will become apparent later in the thesis (i.e., General Discussion), where the AB phenomenon is framed under an attentional shift account. 6 Interference Model The interference model5 is a late selection model based on the theories of Bundesen (1990), and Duncan and Humphreys (1989). Shapiro et al. (1994) argues that limited attentional resources means that only a few visual items are admitted into visual short-term memory (VSTM). In an RSVP experiment, observers were told in advance what the target defining feature(s) are. An internal template defining the target is then constructed, which selects visual items based on task requirements. As the items stream past, the visual system assigns a weight to each, which determines whether it enters VSTM. When an item is assigned higher weights, more attentional resources are allocated to it, increasing the likelihood that it enters the VSTM. The assignment of weights to an item is determined by: (a) its match with the preset internal templates of both the targets, such that a higher weight is assigned when the degree of match is higher (i.e., when the distractor highly resembles the target); (b) its temporal contiguity to either T1 or T2, such that items succeeding T1 or T2 is assigned more weights; or (c) its position in the RSVP stream, such that earlier items are allocated more weights. McLaughlin et al. (2001) pointed out that in that typical RSVP stream, four items are likely to be admitted into VSTM. They are (a) T1; (b) the lag 1 distractor; (c) T2; and (d) the post-T2 distractor. Interference occurs when the distractors in VSTM are inadvertently assigned high weights. In this case, more attentional resources are 5 The interference model (Isaak et al., 1999; McLaughlin et al., 2001; Raymond et al., 1995) is also know as the retrieval-competition model (Maki et al., 1997), the similarity theory (Jolicœur, 1998), and the competition hypothesis (Seiffert and Di Lollo, 1997). In order to reduce confusion over the usage of terms, I shall use the term “interference model” when referring to this model. 7 allocated to these highly weighted distractors. As a result, their identity are retrieved at the point of reporting, causing an AB. According to Shapiro et al. (1994), no AB effect manifests when a long interval separates T1 and T2 (i.e., > 500ms). In this situation, both the T1 and lag 1 distractor would have been “flushed out” of VSTM when their initial weights would have returned to zero with time. As distractors similar to the target match the internal template, the interference model predicts that they would receive more weights. This results in more resources allocated to them, increases their interference in VSTM and thus increases the magnitude of the AB. Raymond et al. (1995) manipulated the featural and spatial similarity of the lag 1 distractor with respect to T1, T2 and the post-T2 distractor. They found that AB was attenuated when the lag 1 distractor was dissimilar from the other three critical items.6 The interference model also predicts that the majority of T2 errors should come from the three critical items (i.e., T1, lag 1 distractor, post-T2 distractor). T1 is assigned high weights because of its match with the internal template, while both the lag 1 and the post-T2 distractors are assigned high weights due to their temporal contiguity with targets. In an error analysis (Isaak, Shapiro & Martin, 1999), T2 errors were shown to be non-random. Isaak et al. also manipulated the number of competing letter distractors in the RSVP stream and found that AB magnitude increased and T2 sensitivity declined as the number of letter distractors increased. This suggest that the presence of interfering distractors, especially the three critical distractors (i.e., T1, Lag 1 distractor, post-T2 distractor), modulated the AB effect. 6 Similar effects were observed in Chun & Potter’s (1995) study, although they proposed a different account for the observed effects. This will be described in a later section. 8 Why then does the blank modulate the AB? Shapiro et al. (1994) argued that as a blank is highly dissimilar to both T1 and T2 (i.e., the blank contains no features), it will be assigned small or no weight. Thus, the blank would not compete with the other items for attentional resources, resulting in the attenuation of the AB. One might even argue that the blank would not enter into VSTM as an “item”. Thus, it cannot interfere with T2 retrieval from VSTM, and this allows T2 to be reported without errors. However, Grandison et al. (1997) demonstrated that a “blank”7 inserted into lag 1 caused an AB. They claimed this finding cannot be reconciled with the interference model. They argued that a blank in lag 1 attenuates AB not because it is assigned no weight, but because the blank would fail to mask T1. Grandison et al.’s explanation supported the two-stage processing model proposed by Chun and Potter (1995), which is described below. Processing Model The central claim of the processing model (Chun and Potter, 1995)8 is that the AB effect is caused by a processing bottleneck. Chun and Potter claimed their model “extends Broadbent and Broadbent’s (1987) observations that early stages of 7 The blank condition in Grandison et al.’s (1997) study was slightly different from Raymond et al.’s (1992) blank condition as it was a blank screen flash where the luminance value of the entire screen changes. 8 Others have called this account the processing bottleneck model (McLaughlin et al., 2001), the perceptual-interference model (Maki et al., 1997), and the delay-of-processing hypothesis (Seiffert & Di Lollo, 1997, while other researchers have modified the original Chun & Potter (1995) model so that the results of their experiments fit the general model specification. In order to provide a clear terminology in discussing these variants of two stage processing models, I shall use the term “two stage processing model” to refer to Chun and Potter’s original account, but use the term “processing model” to refer to all variants of the two stage processing models (e.g., McLaughlin et al., 2001). 9 detection are succeeded by more demanding and capacity-limited processes (and) this type of two-stage conceptualization dates back to Neisser’s (1967) proposal that preattentive processes guide the operation of a focal attention stage” (p. 122). Giesbrecht and Di Lollo (1998) extended it to incorporate visual masking by the object substitution phenomenon (Enns & Di Lollo, 1997, 2001). Jolicœur (1998, 1999a, 1999b; Crebolder, Jolicœur & McIlwaine, 2002; Jolicœur & Dell’Acqua, 2000) proposed a central interference theory that focused on the limitations of response selection. He claimed that “the two stage model is a special case of the central interference theory” (Jolicœur, 1998, p. 1028). Chun and Potter’s processing model assumes two processing stages. The first stage is similar to the preattentive stage in various theories of spatial and temporal selective attention (e.g., Hoffman, 1978; Shiffrin & Gardner, 1972; Treisman & Gelade, 1980; Wolfe, Cave & Franzel, 1989), where the features of all stimuli are extracted. However, sensory representations formed in this stage are transient. Unless they receive further processing and consolidation, they are subjected to rapid degradation and forgetting. Items possessing target attributes (e.g. colour, letter case, semantic category) are flagged. They then undergo further processing in a second stage, where they are consolidated in VSTM. Otherwise, their sensory representation will degrade and their identities unrecoverable. Chun and Potter proposed that the second stage only commences with the detection of target feature in the first stage. That is, the second stage is initiated by a transient attentional response signaling target appearance. This lasts approximately 100ms (Nakayama & Mackeben, 1989; Weichselgartner & 10 Sperling, 1987). The timing and resolution of this transient response results in the lag 1 distractor also entering the second stage due to its temporal contiguity with T1. During the second stage, the target is identified and consolidated in VSTM. Any distractor (i.e., noise) information in the VSTM is discarded. However, this second stage is assumed to be capacity-limited. Thus, the number of items that can enter it at any point of time is limited (i.e., 1 to 2 items). While the second stage is occupied, all other items flagged as potential targets in the first stage are denied entry. This means that they are not processed beyond the first stage. A target that does not undergo processing in the second stage would not be consolidated in VSTM. This means its visual code will degrade, hampering its recovery, which results in the AB. The crux of the processing model is the amount of time T1 processing is prolonged in the second stage. In other words, T1 processing difficulty is an important factor. As processing difficulty of T1 increases, the time required for its consolidation in the second stage should also increase. This delays T2 processing further, producing a larger AB effect. Recall the interference model (Shapiro et al., 1994) predicts that increasing target-distractor similarity results in a larger AB. The processing model makes a similar prediction, but attributes the increased AB effect to increased T1 processing difficulty. When T1 and the succeeding lag 1 distractor are highly similar, selecting the target for consolidation in the second stage would be more difficult, increasing the second stage’s processing time for T1 (Chun & Potter, 1995). Chun and Potter’s original definition of “processing difficulty” is conceptualized at the semantic-level, such that it is high-level masking (i.e., semantic similarity between items) that increases processing difficulty. However, the 11 manipulation of semantic similarity in their study (Experiments 4 & 5) is confounded with low-level masking (i.e., sensory masking), such that high-similarity distractors (i.e., digits) had a higher masking effect than low-similarity distractors (i.e., keyboard symbols). Other researchers (e.g., Grandison et al., 1997; Seiffert & Di Lollo, 1997) extended the idea of difficulty to low-level processing. Seiffert & Di Lollo argued that the backward masking effect of the lag 1 distractor on T1 is also increased when both are similar.9 According to this account (e.g., Breitmeyer, Ehrenstein, Pritchard, Hiscock & Crisan, 1999; Grandison et al., 1997; Seiffert & Di Lollo, 1997), the backward masking effect on T1 degrades it, making it more difficult to process. Grandison et al. demonstrated that a low-luminance blank with less masking properties attenuates the AB effect more than a high-luminance blank. Seiffert and Di Lollo demonstrated that a blank in lag 1 attenuates the AB effect less when T1 is masked by a spatially overlapping or lateral distractor in the T1 frame. In a crucial experiment, Grandison et al. demonstrated that even when the semantic category of T1 and the lag 1 distractor is highly dissimilar (T1 = letter, lag 1 distractor = colored blank), an AB was produced when the lag 1 distractor had low-level masking properties. These findings support the low level masking account, and reject the argument that high-level masking increases processing difficulty.10 The processing model accounts for the blank in lag 1 in the following way. T1 processing will be greatly facilitated as the blank does not mask T1. When T1 is not masked, its visual code will not be degraded and this enhances its processing (e.g., 9 Although target-distractor similarity increases masking effect, it is not a pre-requisite for masking to occur (Grandison et al., 1997). 10 As I have rejected the high level masking account, all future references of “masking effects” in this thesis will refer to low level masking effect unless otherwise stated. 12 Grandison et al., 1997; Seiffert & Di Lollo, 1997). This means that T1 processing will be completed faster, and decreases the probability that the T2 code degrades while waiting for entry into the second stage. Hence, T2 would be better recovered in the second stage, leading to an attenuation of the AB. Comparisons Between Processing and Interference Models It is probably not incorrect to say the processing model is better supported in the literature than the interference model. Many researchers (e.g., Jolicœur, 1998; Giesbrecht & Di Lollo, 1998; Seiffert & Di Lollo, 1997) have interpreted their data in the context of the processing model. However, these data do not necessarily contradict the interference model, as several common aspects of the AB can be equally accounted for by both models. Two such examples highlighted above include: (a) the effects of targetdistractor similarity; and (b) the attenuation of the AB when a blank is inserted in lag 1. Both models also account for the lag 1 sparing effect equally well. According to the interference model, this effect occurs when T1 and T2 are contiguous because there is no intervening distractor that garners the same resources. Hence there is no (or less) interference of T2 retrieval from VSTM. The processing model argues that when T2 appears in lag 1, it enters into the second stage along with T1 because of its temporal contiguity. This is because the transient attentional response that initiates the second stage lasts approximately 100 ms (Nakayama & Mackeben, 1989; Weichselgartner & Sperling, 1987), and draws both T1 and the trailing distractor 13 from the first stage into the second stage. Hence, when T2 appears at lag 1, it would be processed together in the second stage. One often cited difference between the two models is that T1 processing difficulty is not a feature of the interference model. While the processing model predicts that processing difficulty modulates the AB, the interference model assumes no relationship between T1 processing difficulty and magnitude of AB. However, the evidence for the relationship between T1 processing difficulty and the magnitude of AB effect is inconclusive. Although Shapiro et al. (1994) and Raymond et al. (1995) found the correlation between T1 processing difficulty and AB magnitude to be nonsignificant11, Seiffert and Di Lollo (1997) claimed this non-significance was due to a lack of power. They performed a different correlation analysis between T1 processing difficulty and magnitude of the AB over a larger sample and found a significant correlation.12 Grandison et al. (1997) also found a significant correlation between T1 processing difficulty and magnitude of the AB effect in their study.13 Using a speeded T1 task, Jolicœur (1998, 1999) found that magnitude of the AB was correlated to 11 Shapiro et al. (1994) employed d’ as an indicator for T1 processing difficulty, while Raymond et al. (1995) employed T1 error rates. AB magnitude in both experiments was quantified by calculating the area above the curve relating percentage correct T2 detection to T2 relative serial position. 12 The studies from which Seiffert and Di Lollo sampled were: (a) Seiffert and Di Lollo, 1997; (b) Raymond et al., 1992; (c) Shapiro et al., 1994; (d) Raymond et al., 1995; and (e) Chun and Potter, 1995. T1 identification accuracy was employed as an indicator of T1 processing difficulty, while AB magnitude was calculated by taking the difference between 100% and the mean percentage correct on T2 task at SOA between 180 and 540 ms (200 to 600 ms in the case of Chun & Potter, 1995), and then summing the values. 13 It must be noted that Grandison et al. (1997) only found a significant correlation between T1 processing difficulty and magnitude of the AB effect when they correlated AB magnitude and T1 identification accuracy for all participants. The correlation was not significant when mean AB magnitude and mean T1 identification accuracy for experiments was used. T1 identification accuracy was used to indicate T1 processing difficulty, while AB magnitude was calculated according to Raymond et al.’s (1995) method. 14 reaction time (RT) of T1, such that a longer RT for T1 was associated with a larger AB.14 In the studies cited above, the relationship between T1 processing difficulty and magnitude of the AB effect was analyzed in a post-hoc manner. T1 processing difficulty was inferred from measures such as T1 identification accuracy and T1 RT. McLaughlin et al. (2001) manipulated the perceptual quality of T1 code in order to test the processing difficulty hypothesis. The duration of both T1 and its subsequent mask were varied.15 Although McLaughlin et al. found no relation between T1 identification accuracy and magnitude of the AB16 using a skeletal RSVP paradigm, they drew the conclusion that the findings do not undoubtedly favor either the interference or processing model. But they argued their findings place significant constraints that require modifications from both models.17 Ward et al. (1996, 1997) manipulated various aspects of T1 processing difficulties and demonstrated no relationship between T1 processing difficulty and magnitude of the AB effect. However, they did not couch their findings in terms of 14 In this case, RT of T1 is the indicator of T1 processing difficulty, while the AB magnitude was represented directly by P(T2|T1) identification accuracy. 15 The ISI between T1and the lag 1 distractor was kept constant at 15 ms, while the summed duration of T1, ISI and the mask was 105 ms. 16 T1 processing difficulty was indicated by T1 identification accuracy, while AB magnitude was calculated using a formula derived by McLaughlin et al. (2001). It must be noted that McLaughlin et al. did not directly establish the correlation between T1identification accuracy and magnitude of the AB effect for an RSVP stream paradigm, but instead chose to correlate the performance of the skeletal RSVP and RSVP stream task (Experiment 3) to indirectly infer the relationship. 17 Although the non-relationship of the T1 identification accuracy and magnitude of the AB effect supports the interference model, McLaughlin et al. (2001) argued the fact the magnitude of the AB effect is similar for the different conditions of difficulty is contradictory to the predictions of the interference model. This is because the longer presented T1 (easy condition) should receive more weights, which leaves a smaller amount of weights for T2 and result in a larger AB effect. Hence, McLaughlin et al. claimed that the interference model need to operationalise the concept of “temporal contiguity” and “weights in VSTM” properly to account for their findings. On the other hand, although the non-relationship between T1 identification accuracy and magnitude of the AB effect do not support the processing model, McLaughlin et al. proposed that the processing model can still be accepted if one assumes that only difficulty at the post-perceptual level should affect the AB. 15 the interference or processing models, but instead proposed the attentional dwell model, which is described next. Attentional Dwell Model Duncan and his associates (e.g., Duncan et al., 1994; Ward et al., 1996, 1997) proposed that attention is a sustained state during which representations of relevant objects become available to guide behavior. This view contrasted with the view that attention is a high-speed switching mechanism. Using a skeletal RSVP paradigm, Ward et al. (1996) found the AB effect did not depend on: (a) perceptual masking; (b) the number of item attributes to be identified; (c) the number of responses made; or (d) the limits in the number of locations that must be attended. In other words, they did not find a relationship between T1 processing difficulty and magnitude of the AB effect.18 However, they found that the AB was dependent on the number of attended items. Hence, Ward et al. proposed a “parallel competitive system determining the allocation of visual processing resources” (p. 106). By this account, items compete in parallel for a share of limited capacity visual-processing resources, according to their match to a target template. This competition resolves gradually over several hundred milliseconds, and the winners engage the visual processing mechanisms at the expense of the losers. Ward et al. claimed that it is this competition that results in a sustained state of attention in which representations of the selected items and all their properties are 18 Ward et al. (1996, 1997) used T1 identification accuracy as an indicator of T1 processing difficulty, while magnitude of the AB effect was calculated directly from P(T2|T1) identification accuracy. 16 available to control behavior. Ward et al. (1997) provided further support for the attentional dwell hypothesis by replicating their basic findings using a RSVP “stream” paradigm19. According to the attentional dwell model, a blank inserted in lag 1 attenuates the AB effect because there is one fewer item to compete for the allocation of visual processing resources. Therefore, the competition that results in the sustained attentional state resolves faster. Using a skeletal RSVP paradigm, Moore et al. (1996) demonstrated that when T1 is unmasked (i.e., it is followed by a blank), the AB effect was in fact attenuated. To explain these results, Ward et al. (1996) claimed that “attended objects appearing within several hundred milliseconds of each other must share some form of visual processing resources, and therefore suffer divided attention costs. The first relevant object presented engages the majority of these resources, and only gradually do these resources become available for other objects” (p. 102). Present Study Shapiro (2001) highlighted one fundamental issue of attention research: the nature of its timecourse. The question the issue engaged is this: “Do we continuously process information or does our processing ability ebb and flow?” (p. 2). The AB suggests the latter is true. The existence of the AB shows that attention is limited temporally, such that when attention selects a target, attentional processing becomes 19 However, there was a difference in for the skeletal and standard RSVP paradigms when T1 and T2 were identical (i.e., both X), with the latter showing an increased T2 interference. However, Ward et al. (1997) accounted for this result within a type-token explanation of the repetition blindness framework (e.g., Kanwisher, 1987). As the present study is not concerned with type-token differentiation, I note this result but will not discuss it in detail. 17 unavailable temporarily for the subsequent targets. In this thesis, I seek to investigate the underlying cause for these temporal constraints. I argue that revealing these constraints would be a necessary step to an understanding the mechanisms underlying attentional control. The extant AB models offer different accounts of the temporal constraints underlying attention. The interference model focuses on the post-perceptual competition amongst items in VSTM. The processing model claims that it is a processing bottleneck, while the attentional dwell model argues that it is the online competition among visual items for limited processing resources. These models may differ substantially, but the findings from the various studies (e.g., Chun & Potter, 1995; Shapiro et al., 1994; Ward et al., 1996) agree on two issues: (a) the manipulation of the lag 1 distractor had the largest modulating effect on the blink, suggesting the locus of the underlying cause for the AB lies at lag 1; and (b) a blank inserted in lag 1 causes the greatest attenuation of the AB, suggesting that understanding the effects of the blank on the blink is crucial to an understanding of the underlying cause of the temporal limits of attention. In this thesis, I argue that the failure to transfer attentional control to a new target is the underlying cause of the temporal constraints of attention (e.g., Chua, 2005). More specifically, I argue that an important factor modulating the transfer of attention control is how effective termination of T1 is signaled to the visual system. The basis of this argument is derived from two important theories: (a) the theory of visual information acquisition proposed by Loftus and his colleagues (e.g., Busey & 18 Loftus, 1994; Loftus et al., 1992; Loftus & Ruthruff, 1994); and (b) the temporal coding hypothesis proposed by Dixon and Di Lollo (1994). In order to facilitate reading of this thesis, a brief overview of the experiments conducted is described here. In Experiment 1, the critical manipulation is inserting a clone of T1 in the lag 1 position. T2 performance for the repeat-T1 condition with respect to both the blank and baseline conditions are predicted for each of the AB models. For the baseline condition, the lag 1 distractor is a randomly chosen letter, while the lag 1 distractor for the blank condition is a blank with a luminance similar to the background. The baseline and blank conditions in the current experiment is highly similar to a typical two target RSVP presentation in the AB literature, and they are described in detail in the Methods section in Chapter 2. The results from Experiment 1 contradict the above described models. In order to accommodate the findings, the attentional engagement hypothesis proposed by Chua (2005) is introduced, where the AB is framed under an attentional shift framework (Posner & Peterson, 1990). It is hypothesized that attention fails to disengage from T1 rapidly enough when T1 termination is not signaled effectively to the visual system. In Experiment 2, three different types of lag 1 distractors are employed to test this hypothesis. The magnitude of T1 termination signal is manipulated by varying the similarity between T1 and the lag 1 distractor. The results from Experiment 2 support the attentional engagement hypothesis. In all these experiments, the lag 1 distractor was systematically manipulated. All experiments included a baseline and blank condition. Both these conditions are 19 critical as they help to discriminate between alternative accounts when compared to the critical condition. 20 Chapter 2 Single-Stream Experiment Experiment 1 In Experiments 1, I seek to differentiate between the interference, the processing and the attentional dwell models by presenting a clone of T1 in lag 1. In Experiment 1a, the luminance of the repeat-T1 item is similar to the target (i.e. 100 cd/m2), while in Experiment 1b the luminance of the repeat-T1 is similar to the distractor (i.e. 20 cd/m2). For all the experiments conducted in this study, the targets were demarcated from the distractors through a luminance difference, such that the distractors were 21 ‘black’ while the targets were ‘white’ in color. As both T1 and T2 are demarcated by similar target defining attributes (i.e. same luminance/color), we can exclude task switching as a probable account of the AB (Kawahara, Zuvic, Enns & Di Lollo, 2003). One might argue from the research on repetition blindness (RB) (e.g., Kanwisher, 1987; Mozer, 1989; Park & Kanwisher, 1994) that inserting the T1 clone at lag 1 will result in participants not perceiving the repeat-T1, and thus treat it as an “unitization” of the actual T1. On this view, one would argue the repeat-T1 might not be processed by participants (i.e. they are not tokenized as a separate object from T1) and thus presumably does not enter VSTM. Thus, the repeat-T1 essentially becomes a “blank”. Assume that attentional deployment on a target. lasts around 200 ms (Sperling & Weichselgartner, 1995). If participants do not perceive the repeat-T1 due to the RB effect, then T2 performance for the repeat-T1 condition should be similar to the blank condition. In fact, one might also argue that this will predict the same effect for the repeat-T1 conditions whether the T1 clone is in target or distractor luminance. However, even if there were an RB effect such that participants were not consciously aware of the T1 clone, this might not affect the AB as the repeated-T1 distractor at lag 1 might also have properties that a normal distractor might have (e.g., masking, distinct object) that modulates the AB effect. In fact, Chun (1997) demonstrated a double dissociation of the AB and RB effects. Hence, there is no reason to expect the manifestation of the RB effect (if any) to have an influence on the AB effect. Given the above considerations, I shall first consider the situation whereby the repeat-T1 distractor is treated as a distinct object from T1 pertaining to the different 22 AB models in the following sections. The situation whereby the repeat-T1 might not be treated as a distinct object due to an RB effect will also be considered in the Results and Discussion section, where I will make the claim that none of the extant AB models can account for the data of Experiments 1a and 1b, regardless of whether the repeat-T1 is treated as a distinct object from T1 or not. Repeat-T1 and the Interference Model As the target is defined by a specific luminance (i.e., 100 cd/m2) in the RSVP stream, this means that the internal target template formed would be described in terms of luminance. Thus, any item in the RSVP stream that matched the internal target template would be assigned a higher weight. Therefore, a higher weight should be assigned to it. In Experiment 1a, the luminance of the repeat-T1 was similar to the target’s. Therefore, the interference model ought to predict a larger AB effect for the repeat-T1 condition when compared to the baseline condition. The repeat-T1 item should be allocated more resources and thus be more effective in interfering. In Experiment 1b, the luminance of the repeat-T1 was similar to distractor luminance. As the target defining attribute is increased luminance (i.e., 100 cd/m2), the weight assigned to the repeat-T1 in distractor luminance should be similar to any other distractors. Thus, for Experiment 1b, the interference model predicts that the baseline and repeat-T1 conditions should produce the same AB. Shapiro et al. (1994) claimed that the interference model is a late selection model. This means the identity of both T1 and the lag 1 distractor is already known 23 before target selection. This raises a concern for the above predictions, as one might argue that no weights would be assigned to the repeat-T1 in both experiments. This is because the visual system might realize the repeat-T1 cannot be T1. In this case, as the repeat-T1 in both experiments would be assigned no weights, the interference model would predict that the repeat-T1 and blank conditions would produce the same AB. Repeat-T1 and Processing Model Using a partial-report cueing procedure, Bjork and Murray (1977) reported that when a distractor was identical to a target, target identification performance was worse. This finding suggested that repeating an identical stimulus led to a greater masking effect. They attributed this to the feature-specific inhibition at perceptual level. Although Bjork and Murray used a lateral masking paradigm,20 I argue that their finding is also relevant for an RSVP paradigm. Bjork and Murray ensured that the repeated stimulus functioned as interfering noise and not redundant signal by making targets spatially uncertain. However, there was no temporal uncertainty of target for Bjork and Murray’s experiment. In this experiment, although the RSVP task ensured spatial certainty of the target, temporally uncertainty was introduced. Thus, as in Bjork and Murray, the repeat-T1 would probably function as masking noise and not as a redundant signal. 20 In Bjork and Murray’s (1977) experiment, a partial-report cueing procedure on a 4 x 4 display matrix is employed. Two separate letters appears in two different columns. The target is cued at the post-mask display after the presentation of the letters, with an arrow pointing to one of the columns where a letter appeared. Bjork and Murray argue this serves two important functions: (a) the noiseletter interference is concentrated at a perceptual rather than a decisional level; and (b) it allows the physically identical letter to be treated as noise rather than redundant signals. 24 Repeating T1, therefore, should result in increased difficulty in T1 processing. Hence, the processing model predicts a larger AB effect for the repeat-T1 condition than the baseline condition in both Experiments 1a and 1b. There should also be a larger AB effect in Experiment 1a than in Experiment 1b. This is because the higher luminance repeat-T1 in Experiment 1a (i.e. 100 cd/m2 vs. 20 cd/m2) would be expected to produce a higher low-level masking effect (e.g., Breitmeyer, 1984; Grandison et al., 1997). Repeat-T1 and Attentional Dwell Model According to Ward et al. (1996), “objects which are a better match to a desired target specification engage limited processing resources more strongly – and possibly for longer – than objects which are a poorer match” (p. 105). In Experiment 1a, one expects the competition between T1 and subsequent items (i.e., lag 1 distractor) for visual processing resources for the repeat-T1 condition to be larger than the baseline condition. This is because the repeat-T1 has the same luminance as the target and will engage limited processing resources more strongly than a normal distractor. For Experiment 1b, one expects the repeat-T1 in distractor luminance to engage limited processing resources only as strongly as a normal distractor. In this case, one expects the repeat-T1 condition to produce a larger AB than baseline in Experiment 1a, and the same AB with the baseline condition in Experiment 1b. Essentially, the attentional dwell model makes the same prediction as the interference model. This is because both models are based on Duncan and 25 Humphrey’s (1989) theory of visual selection. However, the difference between the interference and attentional dwell models is that the former is an offline model while the latter is an online model. Method Experiment 1 employed a two-target RSVP paradigm (See Figure 2). A single trial consisted of a series of upper-case letters presented successively in a fixed location. Each letter subtended a visual angle of less than 1°. For all the experiments, the letters I and O were excluded from the letter sequence. The stimulus onset asynchrony (SOA) was 100 ms, while the inter-stimulus interval (ISI) was 50 ms: the letter appeared for 50 ms followed by a blank of 50 ms, before the onset of the next letter (See Figure 3). Figure 2: Baseline and Blank Conditions of Single RSVP stream 26 Figure 3. Time Course of Stimulus Presentation The luminance of the target and distractor letters was 100 cd/m2 (i.e. “white”) and 20 cd/m2 (i.e. “black”) respectively. Background luminance was 60 cd/m2 (i.e. “light grey”). For the repeat-T1 condition, the repeat-T1 luminance was similar to targets in Experiment 1a and distractors in Experiment 1b, as shown in Figure 4. Figure 4: RSVP Presentation of Repeat-T1 Condition The letters in the RSVP stream were presented on the centre of the screen. The first and last frames of the stream were always a paragraph symbol, each 27 presented for approximately 250 ms. The stimuli for each trial consisted of 24 letters, with I and O excluded. Each letter appeared as a target with approximately equal frequency. For each trial, the two target letters were predetermined, and then removed from the letter set. The order of the remaining 22 letters was randomly shuffled. The position of the first target (T1) was randomly chosen from frames 6-14. The second target (T2) was then inserted in the frame specified by the lag condition. The RSVP sequence ended 4 frames after the second target. For all the experiments, participants were briefed that there were only two targets per trial and the reporting of two letters at the end of each trial was compulsory. A dialog box appeared after the presentation of the RSVP sequence. Participants entered the identities of the two unique white targets in order of appearance. After their response was recorded, the identities of the two targets were revealed, thereby providing feedback. For all the experiments, the main manipulation was the lag 1 item. The main dependent variable is the accuracy of the identification rates of the first target (i.e. P[T1]), and the identification rates of the second target given the first target (i.e. P[T2|T1]). Each experiment consisted of 16 blocks of trials. Apparatus The stimuli for all experiments were generated and controlled by a Macintosh G4 computer, and presented on a 17” (43.2 cm) Sony monitor. Participants viewed the display freely from approximately 60 cm away. The letters were printed in a 48point Helvetica font (subtending an angle of < 1°). 28 Procedure Participants were briefed to identify the two white color targets and ignore all other distractors. Participants were also told that there will always be only two white targets.21 The experiment was conducted in a dark room where the only source of light came from the monitor. Participants completed 16 blocks of trials, where the first block was treated as practice. Its data were not analyzed. Design The two independent variables for Experiments 1a and 1b were: (a) type of lag 1 distractor (normal distractor letter [baseline], repeat-T1, or blank), and (b) the lag between T1 and T2. In Experiment 1a, the lags were 2-4, and 6. In Experiment 1b, the lags were 2-5, and 7. For the baseline condition, lag 1 distractor was a normal distractor letter. For the repeat-T1 condition, the form of the lag 1 distractor was identical to T1. The repeat-T1 was in target luminance in Experiment 1a and distractor luminance in Experiment 1b. For the blank condition, there was no lag 1 distractor. The variables were factorially crossed. Thus in Experiment 1a, there were 12 trials in each block. In Experiment 1b, there were 15 trials in each block. 21 Technically, for Experiment 1a, there were three white targets in the repeat-T1 condition, since T1 was repeated in lag 1. One can also infer from the literature of Repetition Blindness (e.g. Kanwisher, 1987) that when a stimulus is repeated right after its presentation, it will be not be perceived by an observer. This was the case in the present study as participants denied seeing the repeated letter or three white targets in a single RSVP sequence when questioned about it. 29 Participants There were sixteen participants each in Experiment 1a (13 females, 3 males) and 1b (6 females, 10 males). They were undergraduates from the National University of Singapore and they participated to fulfill course requirements. The data of two participants from Experiment 1a, and one participant from Experiment 1b, were excluded due to high error rates (>70% errors). All participants had normal, or corrected-to-normal, vision. Participants were motivated by the promise of a monetary remuneration of $5 if their performance was satisfactory. In all experiments, the criterion for payment was an average identification rate exceeding 75%. Results and Discussion Both T1 and T2 identification rates as a function of T1-T2 lag for all three conditions are presented in Figures 5 and 6 respectively. In both experiments, the T2 identification curves for all condition suggested (a) the AB was manifest, but (b) the AB was attenuated in both the repeat-T1 and blank condition. In the analysis of all experiments, the necessary Bonferrroni corrections were made in all the post-hoc simple effects tests. The discussion for the rest of this section is organized around key features of the results. 30 Figure 5. T1 Performance for Experiments 1a and 1b (a) T1 Performance For Experiment 1a, the right plot of Figure 5 demonstrated clearly that T1 performance for both the blank and repeat-T1 condition was much better. A 3 (Distractor Type: baseline, blank, repeat-T1) x 4 (Lags: 2-4, 6) repeated-measures ANOVA showed an overall effect of distractor type, F (2, 26) = 18.88, p < .001, ηp2 = 0.592. The main effect of lag was not significant, F (3, 39) = 2.53, p > .07, ηp2 = 0.163. The interaction effect distractor type x lag was also not significant, F (6, 78) = 1.32, p > .25, ηp2 = 0.092. Post hoc tests revealed no reliable differences between the 31 blank and repeat-T1 conditions, F (1, 13) = 2.49, p > .13, ηp2 = 0.161. However, there were reliable differences between the baseline and the blank conditions, F (1, 13) = 20.47, p < .002, ηp2 = 0.612, and between the baseline and repeat-T1 conditions, F (1, 13) = 17.80, p < .002, ηp2 = 0.578. The same pattern of results was obtained in Experiment 1b, as demonstrated on the right plot of Figure 6. An analogous 3 x5 ANOVA showed an overall effect of distractor type, F (2, 28) = 20.45, p < .001, ηp2 = 0.594. The main effect of lag was not significant, F (4, 56) = 2.23, p > .07, ηp2 = 0.137. Surprisingly, the interaction effect distractor type x lag was significant, F (8, 112) = 2.44, p < .02, ηp2 = 0.148.22 Post hoc tests revealed no reliable differences between the blank and repeat-T1 conditions, F (1, 14) = 0.651, p > .43, ηp2 = 0.044. However, there were reliable differences between the baseline and the blank conditions, F (1, 14) = 19.76, p < .002, ηp2 = 0.585 and between the baseline and repeat-T1 conditions, F (1, 14) = 22.90, p < .001, ηp2 = 0.621. According to Bjork and Murray (1977), the repeat-T1 should cause a greater masking effect than a normal distractor. Therefore, it should be expected that T1 identification for the repeat-T1 condition to be worst. However, this was not the case. This suggested that the assumption that the repeat-T1 had a greater masking effect (Bjork & Murray, 1977) might be wrong. This assumption laid in the premise that the 22 The significant interaction effect could be due to the anomalous ‘dip’ in P(T1) performance for lag 7 of the baseline condition. In order to ascertain this fact, a 3 (Distractor Type) x 4 (Lags) repeated measures ANOVA was conducted. Although there was still a main effect of distractor type, F (2, 28) = 20.07, p < .001, ηp2 = 0.589, the interaction effect between distractor type and lag was now not significant, F (6, 84) = 0.774, p > .59, ηp2 = 0.052. Clearly, the anomalous result for lag 7 of the baseline condition was responsible for the significant main effect of lag and interaction effect in the previous analysis. However, there is no theoretical reason to presuppose any main effect of lag and interaction effects between lag and type of lag 1 distractor. Hence, I treated the anomalous result here as a random perturbation in the data. 32 repeat-T1 was treated as interfering noise rather than redundant signal (Bjork & Murray, 1977). One might argue that if the repeat-T1 was treated as redundant signal, T1 processing might be eased. This is consistent with the results obtained, where T1 performance was much better than baseline. However, even if this were the case, the corresponding T2 identification data would still argue against a processing difficulty account. This is elucidated in the next section. (b) T2 Performance Figure 6. T2 Performance for Experiments 1a and 1b 33 T2 identification performance for the baseline condition in both Experiments 1a and 1b were analyzed for a linear relationship.23 There was a significant linear relationship for both Experiments 1a and 1b, F (1, 13) = 37.41, p < .001, ηp2 = 0.742, and F (1, 14) = 67.44, p < .001, ηp2 = 0.828 respectively. This linear relationship may be interpreted as evidence of an AB in both experiments. T2 performance for both blank and repeat-T1 conditions were compared to the baseline condition in both Experiments 1a and 1b. In Experiment 1a, a 2 (Distractor Type: baseline, blank) x 4 (Lags: 2-4, 6) ANOVA revealed that T2 performance for the blank condition was significantly better than the baseline condition, F (1, 13) = 52.27, p< .001, ηp2 = 0.801. The distractor type x lag interaction effect was also significant, F (3, 39) = 4.23, p < .02, ηp2 = 0.245. Simple effects test revealed that the difference between the two distractor types was reliable for all lags.24 An analogous ANOVA, comparing the baseline and the repeat-T1 condition, revealed that T2 performance for the repeat-T1 condition was better, F (1, 13) = 12.13, p < .005, ηp2 = 0.483. The distractor type x lag interaction effect was also significant, F (3, 39) = 7.42, p< .001, ηp2 = 0.363. Simple effects tests revealed reliable differences only at lag 2.25 23 As T2 was not inserted in lag 1 in the present analysis, there was not lag 1 sparing effect and one expects a linearly increasing function rather than a U-shaped function. A pilot study conducted with similar experimental parameters demonstrated a lag 1 sparing effect (and hence a U-shaped AB function) when T2 is inserted in lag 1. 24 For lags 2, 3, 4 and 6, the results of the simple effects test were F (1, 13) = 25.57, p < .001, ηp2 = 0.663, F (1, 13) = 19.32, p < .002, ηp2 = 0.598, F (1, 13) = 20.58, p < .002, ηp2 = 0.613 and F (1, 13) = 25.15, p < .001, ηp2 = 0.659 respectively. 25 For the lags 2, 3, 4, and 6, the results of the simple effects test were F (1, 13) = 28.23, p < .001, ηp2 = 0.683, F (1, 13) = 3.13, p > .1, ηp2 = 0.194, F (1, 13) = 0.585, p > .45, ηp2 = 0.043 and F (1, 13) = 0.065, p > .8, ηp2 = 0.005 respectively. 34 In Experiment 1b, a 2 (Distractor Type: baseline, blank) x 5 (Lags: 2-5, 7) ANOVA revealed that T2 performance for the blank to be much better than the baseline condition, F (1, 14) = 62.95, p < .001, ηp2 = 0.818. The distractor type x lag interaction effect was significant for the blank, F (5, 56) = 3.13, p < .021, ηp2 = 0.183. Simple effects test revealed this difference was reliable for all lags, with the exception of lags 4 and 5.26 An analogous ANOVA revealed that the T2 performance of the repeat-T1 conditions was also much better than the baseline condition, and F (1, 14) = 27.04, p < .001, ηp2 = 0.659. However, the distractor type x lag interaction effect was not significant for the repeat-T1 condition, F (5, 56) = 0.592, p > .67, ηp2 = 0.041. These results show that AB for the blank and repeat-T1 conditions were attenuated in both experiments. The processing model predicted that the AB for the repeat-T1 condition to be larger than the baseline condition in both experiments if the assumptions of Bjork and Murray (1977) were correct. However, recall the possibility that the repeat-T1 at lag 1 might not be perceived as a distinct object from T1 due to an RB effect. This meant that the repeat-T1 might be processed as redundant signal along with T1 and not as interfering noise (Bjork & Murray, 1977). This availability of redundant signal should act like a blank inserted into lag 1. Under the processing model, one would predict T2 performance for the repeat-T1 condition should be equivalent or better than the blank condition. Although T2 performance was attenuated in both Experiments 1a and 1b, the level of attenuation was much lower than the blank condition. Furthermore, if the T1 clone were not tokenized due to an For lags 2, 3, 4, 5 and 7, the results of the simple effects test were F (1, 14) = 25.42, p < .001, ηp2 = 0.645, F (1, 14) = 12.72, p < .004, ηp2 = 0.476, F (1, 14) = 3.06, p > .1, ηp2 = 0.179, F (1, 14) = 3.40, p > .087, ηp2 = 0.195 and F (1, 14) = 13.97, p < .003, ηp2 = 0.499 respectively. 26 35 RB effect, the luminance of the T1 clone should not matter. Therefore, T2 performance for the repeat-T1 condition for both Experiments 1a and 1b should be similar. This prediction was not supported by the data. The interference model makes two possible predictions, depending on whether the repeat-T1 was assigned weights. The first prediction was that the AB for the repeat-T1 condition in Experiment 1a would be larger than the baseline condition, while the AB for both conditions would be similar in Experiment 1b (i.e., when the repeat-T1 was assigned weights). The second prediction was that the AB for the repeat-T1 condition would be similar to the blank condition in both experiments (i.e., when the repeat-T1 was not assigned weights). The attentional dwell model predicted that the AB for the repeat-T1 condition to be larger than baseline in Experiment 1a, but no differences in the AB of both conditions in Experiment 1b. These predictions were not supported by the results in Experiments 1a and 1b. Although both the blank and repeat-T1 conditions were attenuated in both experiments, the simple effects tests suggest a different pattern of attenuation. T2 performance for the blank and repeat-T1 conditions were next compared. For Experiment 1a, a 2 (Distractor Type: blank, repeat-T1) x 4 (Lags: 2-4, 6) ANOVA revealed that AB attenuation was weaker for the repeat-T1 condition, F (1, 13) = 23.44, p < .001, ηp2 = 0.643. The distractor type x lag interaction effect was also significant, F (3, 39) = 2.90, p < .05, ηp2 = 0.182. Simple effects tests revealed reliable differences at all lags except lag 2.27 For Experiment 1b, a 2 (Distractor Type: blank, repeat-T1) x 5 (Lags: 2-5, 7) ANOVA revealed that AB attenuation for the For lags 2, 3, 4 and 6, the results of the simple effects test were F (1, 13) = 0.027, p > .87, ηp2 = 0.002, F (1, 13) = 10.24, p < .008, ηp2 = 0.441, F (1, 13) = 30.98, p < .001, ηp2 = 0.704 and F (1, 13) = 13.32, p < .004, ηp2 = 0.506 respectively. 27 36 repeat-T1 condition was also weaker, F (1, 14) = 11.85, p < .004, ηp2 = 0.458. The distractor type x lag interaction effect was marginally significant, F (4, 56) = 2.52, p < .051, ηp2 = 0.153. Simple effects tests revealed marginally significant differences only at lag 2, while other lags showed no reliable differences.28 This showed that the difference in T2 performance between the repeat-T1 and blank conditions in Experiment 1a contrasted directly with Experiment 1b. This is a critical finding in Experiment 1. Specifically, the critical difference in the pattern of attenuation was at lag 2. This suggests that changing the luminance of the repeat-T1 at lag 1 affected T2 recovery at lag 2.29 For Experiment 1a, the critical finding is that for the repeat-T1 condition the T2 performance at lag 2 is equivalent to the blank condition.30 None of the extant models can account for this result. The data suggest that when the repeat-T1 luminance was identical to T1, T2 recovery at lag 2 was enhanced, but not at the other lags. Under the processing model, this implies that T1, the repeat-T1 at lag 1 and T2 at lag 2 all entered into the second processing stage. However, Chun and Potter (1995) assumed that only two items (i.e., T1 and the lag 1 distractor) enters the second stage. As this finding was not found in Experiment 1b, the implication is that the number of For lags 2, 3, 4, 5 and 7, the results of the simple effects test were F (1, 15) = 6.098, p < .027, ηp2 = 0.303, F (1, 14) = 1.34, p >.26, ηp2 = 0.088, F (1, 14) = 1.13, p > .30, ηp2 = 0.075, F (1, 14) = 0.135, p >. 71, ηp2 = 0.01, and F (1, 14) = 5.45, p >.035, ηp2 = 0.28 respectively. 29 However, it must be noted if the simple effect tests were not corrected for Bonferroni adjustments, this less conservative testing would mean thatT2 performance of the repeat-T1 condition will be significantly different from that of the blank condition at lag 7 in Experiment 1b. This would contradict the statement that changing the luminance of the repeat-T1 at lag 1 affected T2 recovery at lag 2. However, it must be noted that: (a) lag 7 is not tested for Experiment 1a, and thus a comparison of lag 7 of Experiment 1b to lag 6 of Experiment 1a might be inappropriate; and (b) one might argue that the AB effect should have ended by lag 7 (with the common assumption in the literature that AB recovers within 500 ms), and thus the significant difference between the repeat-T1 and blank condition at lag 7 in Experiment 1b might not be an artifact due to the manipulations mean to affect T1 recovery. 30 Recall the T2 performance for the blank condition is theoretically the highest for the given set of parameters within an experiment. Hence, the inference is that T2 performance for the repeat-T1 condition at lag 2 is the highest given the set of parameters under Experiment 1a. 28 37 items that can enter the second stage increases when the repeat-T1 in lag 1 has the same luminance as T1. The processing model has no provisions to account for this interpretation of the findings. Under the interference model, the finding that in Experiment 1a T2 performance improved at lag 2 for the repeat-T1 condition implied that no weights were assigned to the repeat-T1 when its luminance was identical to the target’s (Experiment 1a). A plausible explanation could be the identical T1 in lag 1 was not tokenized as a separate item (Chun, 1997; Kanwisher, 1987; Shapiro, Driver, Ward & Sorensen, 1997), and thus no weight is assigned to it. Therefore, any item that appeared in lag 2 now becomes a ‘lag 1 distractor’. This account is different from the previous prediction that no weight is assigned to both the repeat-T1 in both experiments as the identity of both T1 and the repeat-T1 is known to the visual system prior target selection. In the previous account, the repeat-T1 is tokenized as a separate item, but is assigned no (or little) weight because the system does not mistake the repeat-T1 as T1. The difference between both accounts is that no weight is assigned to the lag 2 distractor when the repeat-T1 is tokenized, but weights are assigned to the item trailing the repeat-T1 when it is not tokenized. The account that the repeat-T1 was not tokenized could explain for the findings in Experiment 1a, as the improved T2 performance at lag 2 could be perceived as a “sparing effect”. In Experiment 1b, as there was a luminance change from T1 to the repeat-T1, the latter could be tokenized as a separate item. Therefore, weights would be assigned to the repeat-T1. The interference model would predict that the AB for both repeat-T1 and 38 baseline conditions should have the same magnitude, as the weights assigned to an item was largely due to its luminance. The data do not support this account. Recall the attentional dwell model predicted that the AB effect of the repeatT1 condition for Experiment 1a would be larger than baseline. According to the attentional dwell model, the number of items determines the duration of the dwell. If the repeat-T1 at lag 1 was not tokenized as a separate item, AB should be attenuated. But the attenuation would be for all lags, and not just for lag 2. In other words, the attentional dwell model also cannot account for the T2 performance for the repeat-T1 condition in Experiment 1a, which is the fact that T2 performance was improved only at lag 2 when the repeat-T1 had the same luminance as T1. To summarize, the T2 performance in the present experiment presents difficulty for extant AB models. None of them predicts the repeat-T1 would attenuate the AB effect. An unexpected finding was that the repeat-T1 luminance modulated T2 performance specifically at lag 2. Both the processing and attentional dwell model has no theoretical provisions to explain this result. Although the interference model might be able to explain this finding by assuming that the repeat-T1 was not tokenized, it still has problem accounting for the data in Experiment 1b. As all the models have problems accounting for these findings, a new framework for the AB is introduced in the next section, where the AB is explained in terms of a failure of attentional shift. It is argued that the new framework would account for the data in Experiments 1a and 1b more accurately. 39 Chapter 3 Double-Stream Experiments Wee and Chua (2004; Chua, 2005) proposed an alternative framing of the AB issue. They argued that the blink occurs because “attentional control could not be transferred readily to a new target” (p. 599). They adopted the general framework of Posner and Peterson (1990), which hypothesized three components in an attentional shift: (a) disengagement from the location of prior target; (b) shift to new target location; and (c) engagement to a new target, where target processing ensues. 40 Although the Posner and Peterson’s (1990) framework was originally conceptualized for spatial shifts, Wee and Chua pointed out it can be adopted for a temporal shift scenario.31 Wee and Chua’s conceptualization is highly similar to Sperling and Weichselgartner’s (1995) episodic theory of attention, which uses Posner’s (1980) metaphor of attention as a spotlight. The engagement of attention to a target is likened to the turning on of a spotlight, which allows the visual stimulus occurring under the spotlight to be processed through the enhancement of its signal (e.g., Kastner & Ungerleider, 2000; Yeshurun & Carrasco, 1998, 1999). Wee and Chua (2004; Chua, 2005) claimed that the main factor modulating the AB is the duration that attention dwells when it is engaged by a target. Chua (2005) claimed that target recovery would be affected by a delay in any one of the three components of the attentional shift. Therefore, the longer attention dwells at a previous target, the longer it takes to shift to a new target. As targets in the RSVP stream appears fleetingly, such a delay in attentional shift hampers target recovery. When an item disappears from the screen, its signal does not terminate immediately due to its sensory persistence (e.g., Coltheart, 1980; Sperling, 1960, Loftus & Irwin, 1998). Loftus and Irwin also pointed out that “information can be 31 For visual search paradigms (e.g., Duncan & Humphreys, 1989; Treisman & Gormican, 1988; Wolfe et al., 1989), target and distractors are presented simultaneously with the spatial location of the target uncertain. Compare this with the RSVP paradigm, where the spatial location of the targets is fixed, but the temporal locus of the targets is uncertain. Even though presentation rate of items is 100 ms per frame for the typical RSVP experiments, targets and distractors still appear as distinct objects (Wee & Chua, 2004). In this case, target selection requires attention to be engaged to the exact temporal locus of an item. In other words, attention must shift to the precise item in the RSVP for it to be deployed, resulting in the processing of the item. 41 acquired from the stimulus for a brief period following stimulus offset in much the same way as it can be acquired while the stimulus is still present” (p. 136).32 In most AB experiments, targets are reported offline, and identification accuracy is emphasized. Targets are presented briefly, and the stimuli appearing before and after each target act as pre- and post-masks respectively. Thus, attention is likely to be engaged at the target for as long as possible to improve the uptake of information acquired, and thus enhance target identification. As information from a visual target can still persists after its offset, attention is likely to remain engaged even when the target has disappeared from screen. The question is when attention will disengage from the target. In this thesis, it is argued that attention would disengage from a target when it detects the target’s termination.33 From this perspective, the AB occurs when the detection of T1 termination lags T1 offset. When the visual system fails to detect T1 termination, it assumes that there is still information available for acquisition, and does not disengage. The late disengagement means that if T2 lags T1 by a short interval, attention is unable to shift immediately to T2. This hypothesis is made under the framework proposed by Wee and Chua (2004; Chua, 2005), which I call the attentional engagement hypothesis. The main thesis is that attention disengages from an item when the visual system detects its termination. As long as the visual system fails to detect T1 termination, attention would not disengage from T1. 32 Coltheart (1980) differentiated between ‘visual persistence’ and ‘information persistence’. In both instances, the stimulus was perceived to be present after its offset, but only information could be obtained from the persisting stimulus in the latter case. In this thesis, all reference to ‘persistence’ (e.g., sensory, visual) actually refers to ‘information persistence’ as defined by Coltheart. 33 A distinction must be made between ‘offset’ and ‘termination’. The former refers to the situation when the item is no longer physically present on the screen, while the latter refers to the situation when the information persistence of the item ends. 42 To detect T1 termination, the visual system must be able to say that T1 has disappeared, and a different letter is on the screen. Dixon and Di Lollo (1994) pointed out that when processing sequences of rapidly varying stimuli (such as the RSVP stream), the visual system has to reconcile two conflicting requirements: (a) to maintain perceptual contiguity, such that temporally separate events that belong together are integrated into a unified percept; and (b) to detect rapid changes, such that closely spaced temporal events that somehow belong apart are segregated from each other. Under their temporal coding hypothesis, the temporal relationship between contiguous stimuli is coded as co-extensive when the visual system hypothesize that the object has not changed or coded as disjoint when the visual system hypothesize that the object has changed. How does this coding process works? Before the mechanisms of this coding process could be outlined, some theoretical ideas developed by Loftus and his associates (e.g., Busey & Loftus, 1994; Loftus, Duncan & Gehrig, 1992; Loftus & Ruthruff, 1994) are introduced below, as they are highly relevant to the temporal coding hypothesis (Dixon and Di Lollo, 1994). According to Loftus and his associates, the visual system acts as a low-pass linear temporal filter on the visual stimulus when it appears. The end result is a sensory response function that relates the magnitude of (neural) activation over time. The visual system is hypothesized to acquire information from this sensory response function. According to Dixon and Di Lollo (1994), the coding process that determines whether temporally contiguous stimuli are integrated or segregated entails correlating the sensory response functions of sequential time-slices. They noted that the temporal 43 coding process is not based on the correlations of the physical stimuli (i.e., raw pixels), but on the correlations of the sensory response function. However, both the physical stimulus and the sensory response functions are thought to be highly related, such that highly similar stimuli would produce highly similar sensory response functions, and vice-versa. When the correlation of consecutive time-slices is large, the visual system assumes that the object has not changed. But when the correlation is low, the visual system assumes that something has changed. In other words, the visual system is more likely to detect that T1 has changed when the item that trails T1 is highly dissimilar (e.g., a blank, a geometric shapes). In this situation, the correlation would be low. When this happens, T1 termination is signaled to the visual system, and attention disengages as a result. In an RSVP stream where items are constantly being replaced, the advent of a new object signals the termination of an old object. Thus, when the visual system detects that the object has changed (i.e., a new object has arrived), T1 would be deemed to have terminated. According to Dixon and Di Lollo (1994), the visual system continuously estimates the correlations, such that “the temporal coding process maintains information about the past history of the visual responses in the form of running averages derived from the samples of visual activity” (p. 50). In this view, it is not just the correlation between the immediate two sequential stimuli that matters, but also the correlation of the previous stimuli. Consider three items, O1, O2 and O3 which are presented in rapid sequence. By the above conceptualization, the correlation of the sensory response function between O2 and O3 is also influenced by the correlation between O1 and O2. This idea would be discussed in detail at a later 44 part of this chapter when the findings of Experiment 1 is explained using the attentional engagement hypothesis. Figure 7. Schematic Correlations of an RSVP stream To illustrate the temporal coding concept, a schematic diagram of the correlations in an RSVP stream is shown in Figure 7. In the diagram, the letters from left to right denotes the order of presentation in the RSVP stream, such that letters left of T1 (i.e., R) denotes the pre-T1 distractors, while letters right of T1 denotes the post-T1 distractors. The arrow on top each pair of letters (e.g., Corr Y1) denotes the raw correlations between the stimuli. The “Hi” and “Low” labels below each letter denotes how well their sensory response function is thought to correlate, where “Hi” means a highly correlation, and “low” means a low correlation. In this thesis, there are two types of changes for the stimuli in the RSVP streams: (a) luminance and (b) shape. Although both contribute in producing the sensory response function, luminance is probably more important when considering the correlations between sensory response functions (e.g., Loftus et al., 1992; Busey & Loftus, 1994), such that the correlation of the sensory response function between 45 two stimuli with similar luminance but different shapes is likely to be higher than two stimuli with similar shapes but different luminance. Letters such as “E” and “F” are likely to be highly correlated as they share many similar letter features. On the other hand, letters such as “F” and “A” are likely to be less highly correlated, as they share less letter features together. From Figure 7, this means that Corr Y1 > Corr Y2. However, due to the random fluctuations of neural noise within the visual system, it has to disregard imperfect correlations between time-slices. In other words, when the correlation between the sensory response functions of sequential time-slices is moderately high enough, the visual system may code them as integrated even if this might not necessarily be the case. When this happens, the visual system would not be able to detect a change immediately, and attention remains engaged to attempt to acquire more information. Of course, the visual system would definitely detect that an object has changed even without a very low correlation value over time. Numerous moderate correlation values over time would still signal an object change. The question is whether the visual system is able to detect a change immediately, such that attention can disengage rapidly. As T1 appearance is demarcated by a change in luminance, its correlation with the preceding distractor is low (i.e., Corr Y1 ≈ Corr Y2, but Corr Y3 is much lower). Hence, the visual system detects a change, and T1 appearance is signaled to it. T1 also correlates lowly with the lag 1 distractor. This should indicate that T1 has terminated. However, sometimes the visual system fails to detect these changes, due to random noise fluctuations in the visual system. Furthermore, given the fact that (a) the correlation between T1 and the lag 1 distractor is required to determine whether 46 something had changed, and (b) time is required to compute the correlations (Dixon & Di Lollo, 1994), it is highly likely that T1 termination is signaled to the visual system after the onset of the lag 1 distractor, such that the sensory response of the time-slice containing the lag 1 distractor could be correlated with that of the timeslices of the previous stimuli. In other words, attention is likely to disengage at lag 1. In other words, attentional disengagement is inevitably delayed, which explains the AB. Consider the next three cases: (a) the lag 1 distractor is a blank, (b) it is highly dissimilar to T1, and (c) when it is highly similar to T1. When a blank is inserted in lag 1, the sensory response function falls to zero (or close to zero). This means that the correlation would also fall to zero, which is a good signal that T1 has terminated. When the lag 1 distractor is highly dissimilar to T1 (e.g., a non-letter), a sensory response function is produced, though it would be lowly correlated with that of the time-slice containing T1. Because of the low correlation, the signal of T1 termination to the visual system should be relatively good. However, it would not be as strong as compared to a blank. Thus, attention does not disengage rapidly in this case when compared to a blank. When the lag 1 distractor is highly similar to T1 (e.g., a letter), the sensory response function elicited would be highly correlated with that of the time-slice containing T1. Hence, T1 termination would not be signaled strongly to the visual system. In other words, compared to a blank and a highly dissimilar lag 1 distractor, attention would disengage much slower in this case. AB attenuation of the manipulated condition (i.e., blank, repeat-T1) is always inferred through the comparison with the baseline condition, which is a random letter 47 (“I” and “O” excluded) inserted at lag 1. For all the trials in this thesis, T1 onset and offset was always marked by a luminance change, such that it becomes a constant between conditions. Thus, though luminance change should be more important than shape change within a single trial, it is likely that the shape change accounts for the difference in T2 performance between the manipulated and the baseline conditions across trials. The exception to the argument that luminance change is constant between conditions is the repeat-T1 condition in Experiment 1a, where T1 offset was not marked by a luminance change (i.e., the lag 1 distractor was in target luminance). In Experiment 1a, it was found that T2 was better recovered at lag 2 for the repeat-T1 condition. This finding can be explained within the attentional engagement hypothesis. For the repeat-T1 condition in Experiment 1a, not only was T1 offset not marked by a luminance change, it was not marked by a shape change as well. When this is the case, the correlation between time-slices would be very high (i.e., close to 1.0), and T1 termination is unlikely to be signaled to the visual system. Hence, attention does not disengage. When T2 appears at lag 2, although there is a shape change (i.e., T1 and T2 are always different), there is no change in luminance. Recall that luminance change is more important than shape change when considering the correlations of the sensory response function between time-slices. Therefore, correlation between the repeat-T1 and T2 at lags 1 and 2 respectively is also likely to be high. In this case, the signal of T1 termination remains poor, and attention continues to be engaged. Hence, T2 is likely to be recovered along with T1 and the repeat-T1 in a single attentional episode, which results in better T2 performance at lag 48 2. T2 performance is not improved when it appears in other lags because the change in luminance as a result of the intervening distractors between T1 and T2 causes attentional disengagement, such that T1 and T2 are not recovered within the same attentional episode, and T2 recovery is compromised because of the necessary attentional shift. Compare this to the repeat-T1 condition in Experiment 1b, where T1 offset was also marked by a luminance change. Even though there is no shape change, the correlation between the time-slices would be lower when compared to Experiment 1a. Thus, attention is more likely to disengage as a result. When T2 appeared at lag 2, it is highly unlikely that it is recovered along with T1 in a single attentional episode. Thus, T2 performance at lag 2 is not improved as a result. However, one would expect the correlation between T1 and the repeat-T1 in Experiment 1b to be higher than that between T1 and another letter (i.e., baseline condition). This is because there is no change in shape for the former. In other words, T1 termination should be signaled more strongly for the baseline condition compared to the repeat-T1 condition, and attention is likely to disengage much earlier as a result. However, in Experiment 1b, T2 performance for the repeat-T1 condition is intermediate between the baseline and blank conditions. This seems to contradict the idea that the correlation between the sensory response functions of the time-slices modulates attentional disengagement. On the other hand, recall that it is not just the correlation between the immediate two sequential stimuli that matters, but also the correlation of the cumulative response function of the previous stimuli. Suppose O1, O2 and O3 49 represents the sensory response function of the time-slices produced by T1, the lags 1 and 2 distractors respectively. When the correlation between O1 and O2 is higher, as in the case when the lag 1 distractor is a repeat-T1 compared to another letter (i.e., repeat-T1 compared to baseline condition), it is likely that the correlation between the cumulative response function of both O1 and O2 (i.e., O1 + O2) to be much lower than O3. Consider the following analogy. Two similar objects both make a rotation of the same degree (i.e., x°) over the same amount of time (i.e., t). For one of the object, it rotates at a constant speed, such that if the time taken is divided into three equal intervals (i.e., t/3), it makes a rotation of x/3° per interval. For the other object, it rotates at an uneven speed, such that if the time taken is again divided into three equal intervals, it makes two small rotations for the first two time intervals, say x/10° each, and one large rotation (i.e., 8x/10°) for the third time interval. Comparing the third time interval for both objects’ rotations, one would expect the sudden large rotation within the same time interval (i.e., 8x/10° per t/3) to be more distinct than the case where rotation speed was constant (i.e., x/3° per t/3). The “equal time intervals” is akin to the frames of T1, the lags 1 and 2 distractors respectively. The object rotating at a constant speed is akin to the baseline condition, where the correlations between the shapes of T1, the lags 1 and 2 distractors are approximately similar (i.e., moderately correlated due to shared letter features).34 The object rotating at an uneven speed is akin to the repeat-T1 condition, where the correlation between the shapes of T1 and the repeat-T1 is high (i.e., both the first two rotations are small rotations), but 34 The correlation between luminance is not considered here because it was argued that luminance change is a constant between conditions. Recall the argument that shape change accounts for the difference in T2 performance between the manipulated and baseline conditions across trials. Thus, only the correlation of shape is considered here. 50 the correlation between the cumulative response functions of both T1 and the lag 1 distractor is low (i.e., the third larger rotation is distinctively different from the first two small rotations). The argument is that this low correlation between the cumulative response function of the time-slices containing T1 and the repeat-T1 with the sensory response function of the time-slice containing the lag 2 distractor causes T1 termination to be signaled much better to the visual system in the repeat-T1 condition compared to the baseline condition in Experiment 1b. This allows for a faster disengagement from T1, and T2 is more likely to be recovered as the attentional system can engage onto T2 much faster when it appears. Going by the above logic, one might point out a discrepancy in the data of the repeat-T1 condition in Experiment 1a. When T2 is not in lag 2, the correlation between the cumulative response function of both T1 and the repeat-T1 (i.e., luminance = target) and the lag 2 distractor in Experiment 1a should be much lower than the cumulative response function of both T1 and the repeat-T1 (i.e., luminance = distractor) in Experiment 1b. This is because the correlation between T1 and the repeat-T1 is higher in Experiment 1a. Thus, one would expect T1 termination to be better signaled at lag 2 in Experiment 1a, which meant that T2 performance at the later lags (i.e., lag 3 and beyond) should be better than Experiment 1b. The data seem to contradict this, as T2 performance from lag 3 beyond for Experiment 1a is similar to the baseline condition, while that of Experiment 1b is better than baseline condition. In order to reconcile this apparent discrepancy, the argument made by Chua (2005) that the amount of information available for acquisition modulates the amount of dwell time is presented. Recall that the visual system is hypothesized to acquire 51 information from the sensory response functions. Loftus and Ruthruff (1994) claimed the total amount of information available for acquisition is the area under the sensory response function, which they claimed is proportional to the energy of the input stimulus. Chua claimed that when the area of the sensory response function is larger, the time required for the acquisition of information would also lengthen. Busey and Loftus (1994) assumed a threshold value for any sensory response function, where information acquisition would stop when the sensory response function drops below this threshold. Hence, when the area of the sensory response function is larger, the time taken to drop below this threshold would be longer. In Experiment 1a, the repeat-T1 in target luminance meant that the more information is available for acquisition than the repeat-T1 condition in Experiment 1b, where the repeat-T1 is in distractor luminance. Furthermore, Dixon and Di Lollo (1994) pointed out that highly correlated sensory response function (i.e., such as the repeat-T1 condition in Experiment 1a) are likely to be coded as co-extensive, whereby a composite sensory response function with a larger area is formed (e.g., Busey & Loftus, 1994). In other words, attention fails to disengage at lag 2 because of the larger amount of information still available for acquisition, which might retard the signal of T1 termination. When information is still available for acquisition, this might imply to the visual system that the target has not terminated. 52 Experiment 2 In a single-stream RSVP experiment, an item always appeared in the same spatial location after the lag 1 distractor. In this case, the sensory persistence of the lag 1 distractor is likely to be overlapped with the visual code of subsequent distractors. When this happens, the correlation between the sensory response function of time-slices after lag 1 is unlikely to be low. This makes T1 termination poorly signaled to the visual system. As a result, the effect of the manipulation of the lag 1 distractor (i.e., which is to manipulate how well T1 termination is signaled to the visual system) on the AB might not be obvious. Furthermore, both targets always appear in the same spatial location, which might obscure the effects of the attentional shift. In order to address the above issues, a double-stream RSVP paradigm was employed in Experiment 2, where (a) both T1 and T2 were located in different streams; and (b) the left stream (location of T1) always terminates after the lag 1 distractor (or after T1, if the lag 1 distractor was a blank). The schematic correlation between stimuli for the single and double stream experiments is illustrated in Figure 8. Similar to Figure 7, the letters from left to right denotes the order of presentation in the RSVP stream, such that letters left of T1 (i.e., R) denotes the pre-T1 distractors, while letters right of T1 denotes the post-T1 distractors. The arrow on top each pair of letters (e.g., Corr Y1, Corr X1) denotes the raw correlations between the stimuli. It must be noted that any sensory response function produced includes all stimuli onscreen for the particular time-slice. 53 In the double-stream presentation, T1 termination is signaled much better in two ways. First, there would be a large increase in the difference of correlations when the left stream terminates at lag 1, as the number of items onscreen drops from two to one item. This causes a correlation between the sensory response function of timeslices of lags 1 and 2 to be much lower than a single-stream presentation. From Figure 8, this means that Corr X3 < Corr Y3. This increases the likelihood that T1 termination (i.e., a change) is detected in the double-stream compared to the singlestream experiment. Second, no item appeared on the left stream after the lag 1 distractor. Thus, there would be no overlapping of the sensory persistence of the lag 1 distractor and with the visual code of any subsequent distractors. When this happens, no more information can be acquired from the left stream (i.e., because there is no longer a visual item in that spatial location), sending a strong signal to the visual system that T1 has already terminated. Thus, attention is more likely to disengage from T1 and its shift is facilitated. The double-stream presentation also introduces a spatial shift element in the experimental trial. By doing so, this places the emphasis on the attentional shift, which makes the effects of the attentional shift on T2 performance more distinct. This would allow a better understanding of the cognitive mechanisms underlying the attentional shift. In Experiments 2, the lag 1 distractor was varied in order to manipulate the similarity between T1 and the lag 1 distractor, such that the correlation between the sensory response function of time-slices between T1 and lag 1 would also be manipulated. When correlation is low, it is more likely that T1 termination is signaled 54 to the visual system. This modulates attentional disengagement from T1, where the attention would disengage when T1 termination is signaled to the visual system. Figure 8: Schematic Correlations for Single and Double Stream Experiment General Method Each frame of the RSVP stream contained 2 letters appearing next to each other. The display was subtended over a visual angle of approximately 5°. The centreto-centre distance of the two letters was approximately 2.0°. The first target was always in the left stream, and the second target was always on the right stream.35 The presentation of the double-stream RSVP sequence was such that both streams appeared together. The baseline and blank conditions are shown in Figure 9. 35 The uncertainty of target appearance in space was therefore eliminated. To the best of my knowledge, no multiple streams RSVP experiments fixed the location of T1 and T2, with the exception of Visser et al. (1999b). However, the purpose of their experiment was to determine if a spatial shift will obliterate lag 1 sparing. I fixed T1 and T2 location in the current experiment in order to magnify possible existing minute effects between conditions. 55 Figure 9. Baseline and Blank Conditions of Double RSVP Stream Other than the baseline and blank conditions, the lag 1 distractor manipulated for Experiments 2a, 2b and 2c were a repeat-T1, a chimeral and a four-dot distractor respectively, as shown in Figure 10. For both the baseline and the repeated T1 condition, the left stream terminated after the presentation of the lag 1 distractor (i.e. the manipulated variable). For the blank condition, the left stream terminated after T1. The letters in each stream were prepared and randomized as in Experiments 1. The major difference is that no paragraph mask appeared in the location of the left RSVP stream after it was terminated. There was only an paragraph mask after the termination of the right RSVP stream. The target identities in both streams were never identical. The luminance of the target letters was 80 cd/m2 (white) for Experiments 2a 56 and 2c, and 100 cd/m2 (bright white) for Experiment 2b, the distractors was 20 cd/m2 (black), and the background was 60 cd/m2 (light grey). Figure 10. Repeat-T1, Four-dot & Chimeral Condition in Experiment 2 Experiment 2a: Lag 1 Distractor = Repeat-T1 ( Distractor Luminance) The repeat-T1 is inserted in lag 1 to test the attentional engagement hypothesis. As the extant models do not conceptualize the AB in terms of attentional shift, they would not predict a different pattern of T2 performance when the repeat-T1 condition is compared to baseline. This is because the repeat-T1 distractor is highly similar to T1 (i.e., identical shape). The interference model would predict that the same weights must be assigned to the repeat-T1 in both Experiments 2a and 1b, while the processing model would predict equivalent masking effects of the repeat-T1 on T1. 57 The attentional dwell model would also predict the same AB effect for both repeat-T1 and baseline conditions, for the number of items presented in both conditions are similar. However, the attentional engagement hypothesis does not make such a prediction, as the critical factor affecting attentional shift (and indirectly AB magnitude) is how well T1 termination is signaled to the visual system. Recall the argument that a repeat-T1 would signal T1 termination much better than baseline condition due to the lower correlation between the cumulative response function of the time-slices containing T1 and the repeat-T1 and the sensory response function of the lag 2 frame. This would predict a better T2 performance for the repeat-T1 condition compared to the blank condition. Method The design was a 3 (distractor type: normal distractor letter [baseline], repeatT1, blank) x 5 (lags: 2-5, and 7) identical to Experiment 1b. Sixteen undergraduates (7 females, 9 males) participated in the experiment under the same motivational condition. The data of one participant was excluded due to high error rates (>70% errors). 58 Results The data are presented in Figure 11 (left plot, P[T1]; right plot, P[T2|T1]). To test whether the repeat-T1 distractor attenuated AB, T2 performance for the repeat-T1 and baseline conditions were compared. A 2 (distractor type: baseline, repeat-T1) x 5 (lags: 2-5, 7) repeated ANOVA analysis showed that T2 performance for the repeatT1 condition was better than the baseline condition, F (1, 14) = 17.85, p < .001, ηp2 = 0.560. However, the distractor type x lag interaction effect was not significant, F (4, 56) = 1.75, p > .15, ηp2 = 0.111. This suggested that inserting a repeat-T1 at lag 1 attenuated the AB for all lags. T2 performance for the repeat-T1 condition was compared to the blank condition next. A 2 (distractor type: blank, repeat-T1) x 5 (lags: 2-5, 7) repeated ANOVA analysis revealed that T2 performance for both conditions were not significantly different, F (1, 14) = 0.336, p > .57, ηp2 = 0.023. The distractor type x lag interaction effect was also not significant, F (4, 56) = 1.31, p > .27, ηp2 = 0.086. A t-test between T2 performance of the repeat-T1 and blank conditions at lag 2 revealed a significant different, t (14) = 2.68, p < .018.36 This suggests that T2 performance was lower for the repeat-T1 condition at lag 2. 36 However, it must be noted that this probability value would fail the Bonferroni adjustments cutoff (i.e., 0.01). Therefore, this result must be interpreted with caution. 59 Figure 11. T1 and T2 Performance of Experiment 2a Discussion As in Experiment 1b, the repeat-T1 in this experiment attenuated the AB. None of the extant models predicted such a finding. This is because the repeat-T1 is highly similar to T1. One would either expect a larger masking effect (Bjork & Murray, 1977), which would result in a larger AB under the processing model. As the repeat-T1 is also a letter, it would be assigned the same weight under the interference model, and AB for both the baseline and repeat-T1 conditions should be similar. The attentional engagement hypothesis would predict for the repeat-T1 condition a better T2 performance than the baseline condition, as the correlation between the 60 cumulative sensory response function of the T1 and repeat-T1 time-slices and the sensory response function of the lag 2 time-slice is lower for the repeat-T1 condition. Hence, the finding in Experiment 2a supports the attentional engagement hypothesis. Recall that the purpose of employing a double-stream presentation was to increase the possibility that T1 termination is signaled to the visual system. This is done by decreasing the number of items onscreen (i.e., termination of the left stream) from two to one. When this happens, the sensory response function of the time-slices containing two items and that containing only one item is lowly correlated. In the blank condition, the left stream terminates earlier by 100 ms (i.e., at lag 1), which means T1 termination is signaled to the visual system earlier. Although compared to the baseline condition, the repeat-T1 also increases the possibility of T1 termination being signaled to the visual system, this is done during lag 2. Recall the argument that the correlation cannot be computed until after the offset of the lag 2 frame, and that time is required for this computation. This probably explains the better T2 performance for the blank condition compared to the repeat-T1 condition at lag 2. Experiment 2b: Lag 1 Distractor = Chimeral Distractor The chimeral distractor is similar to a letter in terms of low-level features as it is comprised of letter features, but differs from a letter in terms of semantic category. A chimeral distractor (See Figure 12) is inserted into lag 1. The chimeral distractors were created by combining two inverted half-letters into a single stimulus. An 61 example is shown in Figure 12, where the chimeral distractor was made up of half an inverted “Q” and “L”. Figure 12. Chimeral Distractor In the context of masking, the chimeral letter provided low-level but not highlevel masking on T1. Therefore, one expects the masking effect of the chimeral distractor to be intermediate between a letter and a blank. However, in terms of signaling T1 termination, the chimeral distractor should be no different from a letter. This is because correlation of shape between stimuli is calculated from their shared features. As the chimeral distractor is created from letter features, one would expect that correlation with T1 for the chimeral distractor to be on the average no different from a baseline distractor. In general, one would argue that the masking effect of the lag 1 distractor is highly correlated with how well it signals T1 termination, as both effects are modulated by shared low-level features (i.e., more shared low-level features leads to larger masking effect and a poorer signaling of T1 termination). In this experiment, T2 was included at lag 1. This is to compare T2 performance between the blank and other conditions at lag 1. According to extant AB models, the blank should improve T2 recovery for all lags, including lag 1. Recall the argument that T1 termination is signaled by low correlations in the sensory response 62 function of time-slices. However, the calculation of this correlation requires time, and cannot be completed until the offset of the sequential item. Thus, the argument is that even though the left stream is terminated (i.e., items onscreen drops from two to one) at lag 1 for the blank condition, attentional disengagement cannot disengage immediately. Therefore, the attentional engagement hypothesis would predict that T2 performance at lag 1 for the blank condition would not be different from the other conditions. Method The design was a 3 (distractor type: baseline, blank, chimeral) x 5 (lags: 1-4, 7) factorial. Target luminance was 100 cd/m2. Seventeen undergraduates (11 females, 6 males) ranging from 18 to 22 years of age participated in the experiment. The data of one participant was excluded due to high error rates (>70% errors). Results The data are presented in Figure 13 (left plot, P[T1]; right plot, P[T2|T1]). To test if the AB for the baseline and chimeral conditions were different, T2 performance for both conditions were compared. A 2 (distractor type: baseline, chimeral) x 5 (lag: 1-4, 7) repeated measures ANOVA revealed no difference in T2 performance between the chimeral and baseline conditions, F (1, 15) = 1.10, p > .31, ηp2 = 0.068. The distractor type x lag interaction effect was also not significant, 63 F (4, 60) = 1.56, p > .19, ηp2 = 0.094. This suggests both conditions produced similar AB. Figure 13. T1 and T2 Performance for Experiment 2b T1 performance for the three conditions was compared. A 2 (distractor type: chimeral, baseline) x 5 (lags) repeated measures ANOVA revealed that T1 performance for the chimeral condition was better than the baseline condition, F (1, 15) = 15.00, p < .002, ηp2 = 0.50. An analogous ANOVA revealed that T1 performance for the blank condition was better than the chimeral condition, F (1, 15) = 6.79, p < .02, ηp2 = 0.312. This suggests T1 performance for the chimeral condition was intermediate between the baseline and blank conditions. 64 T2 performance at lag 1 was compared for the three conditions. A one-way ANOVA (distractor type: baseline, chimeral, blank) revealed no significant differences for the T2 performance for the different conditions at lag 1, F (2, 30) = 0.30, p > .74, ηp2 = 0.02. This suggests T2 performance at lag 1 for all three conditions were similar. Discussion Recall the argument that the masking effect for the chimeral distractor should be lower than a letter. T1 performance indicated that this was the case, as it was better for the chimeral condition than the baseline condition. However, there was no difference in T2 performance. This suggests that the AB is not solely modulated by masking effect. As the amount of masking on T1 is thought to influence its processing difficulty, this finding suggests that processing difficulty does not determine AB magnitude, which argues against the processing model. Although both chimeral and baseline conditions elicited different masking effects, the correlation of the sensory response functions between the time-slices of the T1 and lag 1 frames is postulated to be the same. As T2 performance for both conditions are similar, this suggests that the above correlation, which affects how well T1 termination is signaled to the visual system, explain the T2 performance much better than masking effects. This finding supports the attentional engagement hypothesis. 65 However, one might argue that although the chimeral distractor contains letter features, it is the only non-letter stimulus in each trial, which makes it a singleton (i.e., high-level singleton). Theeuwes (1994) pointed out that attention could be captured by a singleton. When the chimeral distractor appeared at lag 1 at the left stream, attention is likely to be still engaged on T1. Hence, attentional capture by the singleton is likely to delay attentional disengagement from the left stream, which compromises T2 recovery for the chimeral condition. Thus, it is possible that T2 performance for the chimeral condition might have been much better than the baseline condition, which is consistent with the masking argument, but this pattern of result did not manifest because of the chimeral singleton’s capture effect. However, this argument is predicated on the assumption that the identity of each item is known to participant prior selection (i.e., a late selection model). For this experiment, none of the participants reported seeing the chimeral distractor in the RSVP stream. Thus, it is unlikely that the chimeral singleton captured attention. T2 performance for the blank condition was not better than baseline condition at lag 1. The extant models would predict that T2 performance at lag 1 would be better for the blank condition. The interference model argues that no contiguous item trailing T1 would be assigned weights, while the attentional dwell model argues that one less item would compete for visual resources. This would lead to an attenuation of the AB for all lags. The processing model predicts that the blank would not mask T1, allowing it to be processed faster. This should lead to a better T2 performance at lag 1. Furthermore, a pilot study (N=9) (Tan, unpublished data) demonstrated that when T2 was presented simultaneously with T1 (i.e., at lag 0) in a double-stream, T2 66 performance was greatly improved, as shown in Figure 14. This suggests that the failure of an improved T2 performance at lag 1 for the blank condition was not due to (a) the spatial separation of T1 and T2 and (b) processing limitations, for it is possible to attend to and process both T1 and T2 when they appear in different streams. Given the argument of the processing model that items appearing within 100 ms after T1 would be admitted into the second stage, T2 performance at lag 1 for the blank condition should be improved compared to baseline. Figure 14. Data From Pilot Study Depicting T2 Performance at Lag 1 67 Why, then, does the blank not improve T2 performance when it is at lag 1?37 One possible explanation under the attentional engagement hypothesis is that attentional control for the blank condition has not transferred from T1 to T2 by lag 1. Recall the temporal coding hypothesis (Dixon and Di Lollo, 1994) states that the calculation of correlation requires time. When the blank appears on the left stream, and T2 appears on the right stream at lag 1, the correlation between the sensory response function of the time-slice containing T1 with that containing T2 (at lag 1) would be very low. This is due to (a) the number of items onscreen drops from two to one, and (b) the luminance change in the right stream. However, the visual system is likely not to register this low correlation until T2 offset, or at least not immediately upon the onset of the lag 1 frame. In other words, attention is unable to shift immediately upon T1 offset. Why then, does T2 appearing in the right stream at lag 1 not result in its better recovery? If correlation between the sensory response function of the time-slice containing T1 and the lag 1 frames are calculated, then why is it that T2 at lag 1 is not recovered? Recall the calculation of correlation is an ongoing activity. An implicit assumption is that the calculation of correlation between the sensory response function of sequential time-slices and attention are independent 37 It must be noted that Breitmeyer et al. (1999) obtained better T2 performance at lag 1 when a blank followed T1 in a multiple stream (i.e., 4 & 9) experiment. However, their experiment differed from the present experiment in that the “streams” were not true “streams”. At any one time, there were only two items on screen, the target or distractor, and an ampersand mask of the previous item. Thus, all item appearance (including targets and distractor) was marked by an onset transient. This onset transient might have capture attention towards T2 at lag 1, as it has been shown that the onset of a new object always captured attention (Yantis & Hillstrom, 1994). This explains the difference in T2 performance between the blank and baseline condition at lag 1 in their experiment. In this experiment, there is no onset transient (or the "onset" of T2 due to luminance change is already written into the neuronal model (e.g., Solokov, 1960), such that attention does not orient towards it is formed.). Thus, attention would not be captured by T2 onset in this experiment. 68 cognitive processes. Because the correlation is calculated does not necessitate the stimuli is attended. When T1 appears, the low correlation signals its appearance, and attention is engaged onto the spatial-temporal coordinates of T1. In the case when T2 appears together with T1 (i.e., at lag 0), the spatial-temporal coordinates of the attentional engagement would include both streams. However, when T2 does not appear together with T1, the spatial-temporal coordinates of the attentional engagement would only include the left stream. In this case, when T2 appeared at the subsequent lag (i.e., lag 1), attention needs to disengage from T1 before it can shift and engage onto T2. However, this cannot be achieved immediately, as the signaling of T1 termination to the visual system is also not immediate. Thus, the reason T2 performance was not improved at lag 1 for the blank condition might be due to the fact that attentional control is being transferred at that point of time. In other words, the attentional system might be in the process of starting a new attentional episode (i.e., temporal shift of attention). The reason why T2 performance was improved for lag 0 (i.e., when T1 and T2 appeared together) in the pilot study could be due to the fact that both are processed under the same attentional episode (i.e., no temporal shift required). Experiment 2c: Lag 1 Distractor = Four-dot In this experiment, masking effect on T1 was eliminated or largely minimized by inserting a four-dot distractor in lag 1 (Figure 15). Brehaut, Enns & Di Lollo (1999) argued that the lag 1 distractor masks the target through: (a) integration masking (i.e., 69 through contour interaction); and (b) interruption masking (i.e., through the replacing of T1 visual code by the object substitution effect). The four-dot distractor has no overlapping of contours with T1, which eliminates the possibility of integration masking. Enns and Di Lollo (1997) demonstrated a four-dot distractor did not mask a target through the object substitution effect when observers knew where the target was located.38 In this experiment, T1 location is fixed (i.e., it is always on the left stream). One would argue that the object substitution effect would not manifest in this case. As interruption masking is dependent on the object substitution effect, the fourdot distractor is unlikely to mask T1 by interruption. If both integration and interruption masking is essential to producing the AB (e.g., Brehaut et al., 1999; Grandison et al., 1997; Seiffert & Di Lollo, 1997), there should be no difference in T2 performance between the four-dot and blank condition. Figure 15: Four-dot Distractor 38 However, there were studies demonstrating a four-dot distractor could induce a masking effect (Dell’Acqua et al., 2003; Enns & Di Lollo, 1997; Giesbrecht et al., 2003), under the framework of object substitution masking (Enns & Di Lollo, 1997, 2001). There are two conditions for a four-dot distractor to induce a masking effect: (a) target location is unknown prior to its presentation, such that attention has to shift to target location for processing, allowing object substitution masking to occur (Enns & Di Lollo, 1997); and (b) the four-dot distractor is presented after T2 rather than T1 (Dell’Acqua et al., 2003; Giesbrecht et al., 2003), as object substitution masking only occurs when one’s attentional is deficient and unable to engage immediately onto a target. These two conditions are not fulfilled in the current experiment. Hence, I argue the four-dot distractor does not mask T1. 70 The four-dot distractor shares almost no features with and is highly dissimilar to a letter. Hence, the correlation between the four-dot and T1 should be low, and T1 termination is likely to be signaled strongly to the visual system. Thus, compared to the baseline condition, one would predict that the lower correlation for the four-dot condition would lead to a faster attentional disengagement, and a better T2 performance. In fact, one would argue the four-dot distractor is almost similar to a blank. However, despite the fact that both are highly dissimilar to T1 (i.e., non-letter), the four-dot distractor could be categorized as an object while the blank is not an object. Recall the double stream signals T1 termination strongly when the number of items onscreen drops from two to one. Thus, this meant that an earlier T1 termination would be signaled to the visual system for the blank condition than the four-dot condition. From the findings in Experiments 2a and 2b, one would predict that compared to the four-dot condition, T2 performance at lag 2 for the blank condition would be better. Method The design was a 3 (distractor type: baseline, blank, four-dot) x 5(lag: 2-5, 7) factorial. The lag 1 distractor was four dots at the corners of an imaginary square (approximately 1° on each side), as shown in Figure 15. Four-dot distractors were presented randomly in place of letter distractors in both streams before T1 appeared in order to prevent attentional capture by a singleton (e.g., Theeuwes, 1994). With the exception of the four-dot condition where the four-dot distractor was inserted in lag 1, 71 no other four-dot distractor appeared after T1 presentation. Sixteen undergraduates (11 females, 5 males) participated in the experiment under the same motivational condition. The data of two participants were excluded due to high error rates (>70% errors). Results The data are presented in Figure 16 (left plot, P[T1]; right plot, P[T2|T1]). In order to demonstrate the four-dot distractor does not mask T1, T1 performance for the different conditions was compared. A 2 (distractor type: baseline, four-dot) x 5 (lag: 2-5, 7) repeated measures ANOVA revealed that T1 performance for the four-dot condition was higher, F (1, 13) = 47.61, p < .001, ηp2 = 0.785. An analogous ANOVA revealed that T1 performance for the four-dot and blank conditions were similar, F (1, 13) = 0.371, p > .583, ηp2 = 0.024. In order to determine whether the AB for different conditions was similar, T2 performance for the difference conditions was compared. A 2 (distractor type: baseline, four-dot) x 5 (lag:2-5, 7) repeated measures ANOVA revealed that the fourdot inserted at lag 1 attenuated the AB effect, F (1, 13) = 31.69, p < .001, ηp2 = 0.709. The distractor type x lag interaction effect was also significant, F (4, 52) = 3.92, p < .007, ηp2 = 0.232. Simple tests reveal reliable differences at lags 4 and 5.39 For lags 2, 3, 4, 5 and 7, the results of the simple effects test were F (1, 13) = 0.328, p > .57, ηp2 = 0.025, F (1, 13) = 2.77, p > .12, ηp2 = 0.176, F (1, 13) = 25.04, p < .001, ηp2 = 0.658, F (1, 13) = 29.89, p < .001, ηp2 = 0.697 and F (1, 13) = 6.43, p < .025, ηp2 = 0.331 respectively. 39 72 Figure 16. T1 and T2 Performance for Experiment 2c Recall that the masking argument would predict that the AB for four-dot and blank conditions should be similar. Although an analogous ANOVA revealed that T2 performance for the four-dot condition was not significantly different from the blank condition, F (1, 13) = 0.045, p > .83, ηp2 = 0.003, the distractor type x lag interaction effect was significant, F (4, 52) = 3.43, p < .015, ηp2 = 0.209. Simple effects test reveal reliable differences only at lag 2.40 This suggests that the AB for the four-dot and blank conditions were different, such that T2 performance for the blank condition was better at lag 2 than for the four-dot condition. For lags 2, 3, 4, 5 and 7, the results of the simple effects test were F (1, 13) = 10.11, p < .007, ηp2 = 0.438, F (1, 13) = 0.052, p > .82, ηp2 = 0.004, F (1, 13) = 0.001, p > .99, ηp2 = 0.001, F (1, 13) = 2.53, p > .13, ηp2 = 0.163 and F (1, 13) = 2.84, p > .11, ηp2 = 0.179 respectively. 40 73 Discussion According to Brehaut et al. (1999), the blank attenuates the AB because it does not mask T1 through integration or interruption masking. Recall the prediction that the above masking effects would not be found for the four-dot distractor as well. As T1 performance for both the blank and four-dot conditions are similar, this suggests the four-dots do not mask T1. In this case, the masking argument would predict that T2 performance for both the four-dot and blank conditions to be similar. However, compared to the blank condition, it was found that the four-dots impeded T2 recovery at lag 2, which suggest the masking account cannot explain this finding. However, the attentional engagement hypothesis is able to account for the data. Recall that for the blank condition, T1 termination is signaled much earlier than the four-dot condition, as the number of items onscreen drops from two to one earlier by 100 ms (i.e., at lag 1). Although the four-dot distractor is highly dissimilar from a letter, it could still be categorized as an object. Compared to the blank, the signaling of T1 termination via the sudden decrease in the number of items onscreen was delayed. Thus, attentional disengagement for the four-dot condition was also delayed. This accounts for the difference in T2 performance between the four-dot and blank conditions at lag 2. As the four-dot distractor is highly dissimilar to T1, compared to the baseline condition, it would signal T1 termination to the visual system much better, which is supported by the finding that the AB for the four-dot condition was attenuated. 74 However, one cannot completely rule out a masking account in this experiment. It must be noted that the T1 performance for the blank and four-dot condition were at ceiling (i.e., > 99%). Hence, the four-dot might have a masking effect on T1, such as metacontrast or lateral masking (e.g., Breitmeyer, 1984). However, T1 performance did not indicate these possible masking effects as T1 task might be too easy. General Discussion: Experiment 2 Under the attentional engagement hypothesis, when the magnitude of the AB is larger, this implied that attention disengagement from T1 is delayed. As attentional disengagement from T1 is modulated by how well its termination is signaled to the visual system, where attentional disengagement is faster when T1 termination signal is distinct, one would argue that the magnitude of the AB would be smaller when T1 termination is more distinct. Recall that the double-stream paradigm was employed to make T1 termination more distinct to the visual system. Using McLaughlin et al.’s (2001) formula, an AB magnitude for the baseline condition was calculated for each participant. The mean AB magnitude for the single and double-stream experiment were 1.06 (S.E. = 0.116) and 0.74 (S.E. = 0.083) respectively. An independent sample t-test revealed that the AB magnitude of the baseline condition for the double-stream experiments was 75 smaller, t(65) = 2.297, p < .025.41 This suggests the double-stream paradigm was more effective in making T1 termination signal more distinct. For the double-stream experiments, there appeared to be a ‘dip’ in T2 performance at lag 3. A paired sample t-test revealed T2 performance at lag 3 was lower than lag 2 for the baseline condition in the double-stream experiments, t (42) = 2.246, p < .03, MSE = 0.045. This was also found for the blank condition, t (42) = 2.53, p < .015, MSE = 0.034. This suggests the ‘dip’ at lag 3 is reliable. In these experiments, T1 always appeared in the left stream, while T2 always appeared in the right stream. A survey at past AB experiments involving multiple streams (e.g., Breitmeyer et al., 1999; Kristjansson & Nakayama, 2002) revealed no ‘dip’ for T2 performance at lag 3. In these experiments, T1 and T2 locations were not fixed.42 Indeed, in a pilot study, when the location of both T1 and T2 were varied, the ‘dip’ at lag 3 disappeared. Thus, it is when the T1 and T2 locations are fixed that the dip obtained. One plausible explanation for the ‘dip’ in the T2 performance at lag 3 for the double-stream experiments could be due to an improved T2 performance at lag 2. In most AB experiments, T2 performance is worst at lags 2 or 3. Hence, when performance at lag 2 is somehow facilitated, the minima of the lag function appears shifted to lag 3. As the spatial location of T2 in most multiple streams AB 41 One may argue that the AB magnitude cannot be compared across experiments as Experiments 1a and 2b sampled T2 performance from difference lags. Thus, an AB magnitude was derived for each participant from lags 2-4, which were ran in all experiments. An independent sample t-test revealed a smaller AB magnitude for the double-stream experiments, t(65) = 3.025, p < .004, which was the same pattern of result obtained when AB magnitude was calculated from all lags. 42 To the best of my knowledge, the only other multiple RSVP stream experiment to fix both T1 and T2 location was conducted by Visser et al. (1999). However, T2 performance at lag 2 was not collected in their experiment. Hence, there is no way of ascertaining whether fixing target locations in their experiment caused a ‘dip’ at lag 3. 76 experiments is uncertain, participants might develop a strategy of remaining in a state of ‘diffuse attention’ (e.g., Jonides, 1983; Yantis & Jonides, 1990) after T1 has appeared, and not deploy attention onto any stream until the appearance of T2. In this experiment, as the spatial location of both T1 and T2 is fixed, participants probably deployed their attention onto the right stream as soon as attention can be disengaged from T1. The calculation of correlation between time-slices requires time; attention probably could not disengage instantaneously from a previous target within 100 ms. Hence, there is a high possibility that the engagement of attention to the right stream (after disengagement from the left stream) occurs slightly before or at lag 2. When T2 also appears at lag 2, its appearance will coincide with the engagement of attention due to the routine spatial shift, leading to better T2 performance. When T2 does not appear at lag 2, attention probably disengages temporally. Thus, when T2 appears at lag 3, attention might not be able to engage onto it efficiently as it is still in the process of disengaging. The above speculation rests on two assumptions: (a) when there is an attentional engagement due to a spatial shift, there is a temporal attentional engagement that is yoked together with the spatial attentional engagement; and (b) there is an automatic disengagement of temporal attention if the target were not included in the above described attentional episode. If these two assumptions were true, then the ‘dip’ in T2 performance at lag 3 would provide support for the attentional engagement hypothesis, as together with T2 performance at lag 1 for Experiment 2b, it suggests that there is attentional disengagement from T1 and 77 attentional engagement onto T2. However, this explanation remains speculative at best, and more research is required to provide supporting evidence. 78 Chapter 4 General Discussion The set of experiments examined how different T1 distractors modulated the magnitude of the AB. Experiment 1 employed a single-stream RSVP, while Experiment 2 employed a double-stream RSVP. The AB account was framed under the attentional engagement hypothesis (Chua, 2005; Wee & Chua, 2004). The claim is that the AB is due to the fact that attentional control cannot be reassigned promptly to a new target. The manipulation of the T1 distractors was aimed at influencing the 79 similarity between T1 and the lag 1 distractor, such that the correlations of the (a) sensory response function of the time-slices between T1 and the lag 1 distractor, and (b) that of the cumulative response function of T1 and lag 1 time-slices with the sensory response function of the lag 2 time-slice were varied. When sequential items were highly similar, their sensory response functions were also highly correlated. The double-stream RSVP was employed to make T1 termination more distinct. . The main findings are summarized as follows: (a) When the correlation between the sensory response function of time-slices is low, T2 performance was modulated (Experiments 1a and 1b, 2a, 2b and 2c), such that the lower the correlation, the better T2 was recovered. (b) An account based on backward masking cannot explain these data. When masking effects were increased (Experiments 1a and 1b), it did not lead to worse T2 performance. Conversely, decreasing masking effects by using a chimeral distractor did not lead to better T2 performance (Experiment 2b). Also, when masking effects were eliminated or minimized dramatically using a four-dot distractor (Experiments 2c), it still interfered with T2 recovery. (c) AB magnitude for the double-stream experiments was smaller than the singlestream experiments. 80 Repeating T1 at Lag 1 In Experiment 1a (repeat-T1 = luminance of T1), 1b and 2a (repeat-T1 = luminance of distractor), a clone of T1 was inserted in the lag 1 position. Recall Bjork and Murray’s (1977) findings that repeating an identical stimulus resulted in a greater masking effect, and thus should increase the difficulty of T1 processing. However, the T1 identification for the repeat-T1 conditions in these experiments did not suffer. One plausible explanation is that T1 identification accuracy may not a good indicator of processing difficulty.43 Another plausible explanation for the T1 performance for the repeat-T1 conditions was that the assumption that the repeat-T1 increases T1 processing difficulty was not supported empirically. The repeat-T1 might have acted as redundant signal rather than a backward mask. Hence, it could have eased T1 processing (e.g., Jacoby, 1983; Kuwana, 2004). The T1 performance indices in Experiments 1a, 1b and 2a support this assumption. However, whether the repeat-T1 impeded or eased T1 processing would not have affected the argument that processing difficulty did not modulate the AB. Assume T1 processing was facilitated by the repeat-T1 letter. In spite of this, T2 performance for the repeat-T1 condition was not better than the blank. The blank cannot act as a redundant signal. In other words, the blank does not impede T1 43 The improved T1 performance in Experiments 1a, 1b and 2a might be an artifact of experimental procedures. In the repeat-T1 condition, it was possible that processing difficulty was increased. However, this did not reflect on T1 performance as selecting either T1 (i.e. correct response) or the repeat-T1 (i.e. incorrect response) will lead the participants to report the same letter. Even if this were the case, it would make no difference to the arguments of this thesis. The argument is outlined at a later stage of this section. 81 processing, but it cannot improve it. The processing model predicts faster T1 processing for the repeat-T1 than the blank condition. This should result in a better T2 performance for the repeat-T1 than the blank condition. The data contradict this prediction. On the other hand, assume T1 processing was impeded by the repeat-T1. The processing model now predicts delayed T1 processing for the repeat-T1 than the baseline condition. This should result in a worse T2 performance for the repeat-T1 than the baseline condition. Again, the data contradict this. The findings from Experiments 1a, 1b and 2a argue against both interpretations. In other words, regardless whether T1 processing was impeded or facilitated by the repeat-T1, the processing model is not supported. For the interference model, the internal target template is predetermined before each trial (Duncan & Humphreys, 1989; Shapiro et al., 1994). Given that target identities were unknown to participants prior to the start of each trial, the information about letter shape cannot be incorporated into the target template. Only information of target defining attributes would be programmed into the target template, which in these experiments was a specific luminance. Therefore, the match of the repeat-T1 in both Experiments 1b and 2a (luminance = distractor luminance) to the target template should be similar to the other distractors, and they should be assigned the same weight, and thus be allocated the same amount of attentional resources as other distractors. Hence, the interference model would predict similar T2 performance for the baseline and repeat-T1 conditions. The data contradict this. In Experiment 1a, the repeat-T1 would be assigned more weights, and thus obtain more attentional resources. Hence, T2 performance for the repeat-T1 condition should be 82 worse than the blank, which is again contradicted by the data. Thus, the interference model is not supported. The attentional dwell model posits that it is the number of items presented to the participants that determines the magnitude of the AB (e.g., Ward et al., 1996, 1997). For the repeat-T1 condition, the number of items is similar to the baseline condition. Furthermore, the repeat-T1 is of the same category (i.e., letters) as the lag 1 distractor in the baseline condition. Thus, one would expect the competition for visual processing resources for the repeat-T1 to be equally similar to the baseline condition. Hence, T2 performance for the repeat-T1 condition should be similar to the baseline condition. However, the data contradicts this, and the attentional dwell model is not supported. Low-Level Masking and the AB Effect Previous studies have demonstrated the AB effect is modulated by low-level masking effects on T1 (e.g., Breitmeyer et al., 1999; Grandison et al., 1997; Seiffert & Di Lollo, 1997, but see McLaughlin et al., 2001), such that a larger masking effect results in a larger AB effect. This masking effect is commonly framed under the processing model: target processing difficulty is said to increase when masking effects are larger. This delays T1 processing. Consequently, T2 entry into the second stage is also delayed, thus hampering T2 recovery. However, the low-level masking argument cannot explain the findings. Masking effects ought to be larger for the repeat-T1 conditions in Experiments 1a, 1b 83 and 2a (Bjork & Murray, 1977). But they did not result in a corresponding larger AB effect. In fact, the AB effect for the repeat-T1 conditions was attenuated.44 In Experiment 2c, the four-dot distractor should not mask T1 through integration, or interruption masking (Brehaut et al., 1999; Giesbrecht and Di Lollo, 1998). The four-dot distractor did attenuate the AB effect significantly. Now one might argue that this result fit the ideas that the four dots do not mask T1. But T2 recovery was in fact hampered at lag 2 compared to a blank condition. As the blank at lag 1 should also not mask T1, a simple masking argument cannot account for the difference in T2 performance at lag 2. T1 performance for the four-dot condition suggested it did not mask T1, as T1 performance was similar to the blank condition.45 While Experiment 2c demonstrated that similar non-masking effect of the lag 1 distractor led to different T2 performance, Experiment 2b demonstrated the opposite effect. The masking effects of the chimeral and baseline distractors were different, as reflected by the better T1 performance for the chimeral distractor. Yet there was no difference in T2 performance between the baseline and chimeral conditions. These findings suggest that a simple low-level masking account fails to account for the data. 44 However, T1 performance for the repeat-T1 conditions suggests this finding should be interpreted with caution. Masking effect on T1 should be reflected in T1 performance, such that a greater masking effect should lead to lower T1 performance. T1 performance for the repeat-T1 conditions failed to suggest that masking was increased. 45 However, this result should be interpreted with caution, as both T1 performance for both blank and four-dot conditions are at ceiling. It is plausible that the four-dot might mask T1, but not reflected in T1 performance. 84 High-Level Masking and the AB Effect Chun and Potter (1995) argued that semantic categories of distractors can affect T1 processing difficulty. When target and distractor categories are highly similar, T1 processing difficulty is increased. This is because it would be more difficult to select targets from distractors. This should increase the magnitude of the AB. Previous research found that manipulating the semantic category of the lag 1 distractor modulated AB effect (e.g., Chun & Potter, 1995; Raymond et al., 1995). I refer to this effect as ‘high-level masking’. The findings in this thesis do not support the high-level masking account of the AB. In Experiment 2b and 2c, the chimeral and four-dot distractor can be considered as ‘geometric shapes’ which is categorical dissimilar from the target letters. Hence, both should not affect T1 at a semantic level, especially for the fourdot distractor that shared no features with a letter. However, T2 performance for the four-dot condition was similar to baseline at lag 2, which suggests that the four-dot distractor interfered with T1 processing. T2 performance for the chimeral distractor, which was similar to the baseline condition, also contradicts the high-level argument. Therefore, the data does not support a high-level masking account. One might argue that both low- and high-level masking affect T1 processing difficulty. This argument is consistent with the T1 performance of the chimeral distractor. A baseline distractor mask at both low and high levels, but a chimeral distractor only mask at low level. The blank, on the other hand, does not mask at all. The data showed that T1 performance for the chimeral distractor was intermediate 85 between baseline and blank condition. But the same processing argument should predict a corresponding pattern for T2 performance. The data did not support this. Both low- and high-level masking cannot explain the data in this experiment. Therefore, it is difficult to advance the processing bottleneck as a candidate for the mechanism of the temporal limits of attention. A further finding that suggested the processing bottleneck was not an appropriate candidate for the mechanism of temporal limits of attention was that the correlation between AB magnitude and mean T1 performance was not significant, r (216) = -.04, p > .562.46 This suggested that processing difficulty did not modulate AB magnitude in the present thesis. Attentional Engagement Hypothesis When attention is deployed to a target, information acquisition continues until there is adequate information for task performance, or until the visual system detects the termination of the target, as this signals no more information from the target could be obtained. For a typical AB experiment, three factors make it unlikely that adequate information is acquired for task performance with certainty47: (a) the short presentation time of the target (i.e., 50 ms); (b) the masking effects of preceding and subsequent distractors; and (c) the temporal uncertainty of the target. In this case, 46 The correlation between mean T1 performance and AB magnitude calculated from lags 2-4 was also not significant, r(216) = -.038, p > .576. 47 The fact that T1 identification is always high (i.e. at least 85%) seem to contradict the argument that it is unlikely enough information is acquired for task performance with a high degree of certainty. However, one can argue that a high identification rate might not necessary indicate the visual system was 100% certain of target identity. Even though the amount of information acquired is not enough for the visual system to be certain of target identity, sufficient featural information might have been obtained for the system to have a good estimation of target identity. Hence, it is not surprising that T1 errors are not random (e.g., participants mistake E for F). 86 attention continues to be engaged onto the target to increase possibility that more information could be acquired. This happens until target termination is signaled to the visual system, such that information acquisition also terminates and attention is disengaged from the target. According to the temporal coding hypothesis (Dixon & Di Lollo, 1994), when the sensory response functions of sequential time-slices are highly correlated, they are coded as temporally coextensive. When they are poorly correlated, they are coded as temporally disjoint. It must be noted that the correlation per se refers to the sensory response function generated by the visual system for the stimuli, and not the raw pixels of the stimuli onscreen. However, both of them must surely be highly related with each other. Temporally coextensive stimuli are likely to be integrated into a single unified percept (i.e., the visual system senses that target has not terminated yet), while temporally disjoint stimuli are likely to be segregated into separate items (i.e., the visual system senses that target has terminated). When the sensory response functions of sequential time-slices are coded as temporally coextensive, the visual system mistakes the separate stimuli as a continuation of T1 signal. Attention remains engaged for continued information acquisition even though T1 is no longer on the screen. Thus, this delays the (a) attentional shift towards a new target and (b) the attentional engagement onto a new target, which then hampers T2 recovery. On the other hand, it is more likely that T1 termination would be signaled to the visual system if the sensory response functions of sequential time-slices are coded as temporally disjoint. This allows for a faster disengagement from T1, which speeds up attentional engagement onto a new target. 87 Due to random fluctuations in neural noise, the visual system would be unable to tell whether a moderate correlation is due to noise, or the object in view has changed (i.e., target has terminated). Unless the correlation was sufficiently low (e.g., when the lag 1 distractor were a blank), T1 termination would not be signaled effectively to the visual system. Thus, attention might be unable to disengage from a target rapidly enough. When the next target appears within a short time of the previous target, attention is unable to engage onto it. Hence, T2 is not recovered. The relationship between T1 masking effects and T2 performance found in previous AB experiments (e.g., Grandison et al., 1997; Seiffert and Di Lollo, 1997) can also be reinterpreted under the attentional engagement hypothesis. Raymond (1995), and Chun and Potter (1995) demonstrated that target-distractor similarity modulated the AB. As the masking effect of the lag 1 distractor is larger when it is physically similar to T1, other researchers (e.g., Seiffert and Di Lollo, 1997) argued that it is masking that modulates the AB. However, when the target-distractor physical similarity is high, their sensory response function is also highly correlated (Busey & Loftus, 1994). In this case, T1 termination would not be signaled to the visual system strongly. Thus, attention remains engaged, which results in a larger AB. In fact, this argument can account for the global similarity effect found in Chun and Potter’s study (Experiment 4 & 5). 88 Other Studies Supporting The Attentional Engagement Hypothesis The main argument is that within the temporal limits of the RSVP paradigm, attention will continue to dwell at a target as long as possible until (a) sufficient information has been acquired for target identification; or (b) the target has terminated.. Target termination is signaled to the visual system by a large change in correlation between stimuli. Chua (2005) provided support for the first part of this argument by demonstrating that attention dwell is longer when the amount of information available for acquisition is increased by manipulating T1 contrast. It was found that attention dwelled longer when T1 contrast was higher, when more information of the target was available for acquisition. As a higher contrast target is much easier to process than a low contrast target, a higher contrast target should attenuate AB. However, the opposite result was found. Chua argued this provided evidence against the processing model. Instead, the fact that AB dwelled longer when more information was available for acquisition supports the attentional engagement hypothesis. This is also consistent with the finding for the repeat-T1 condition in Experiment 1a, where the repeat-T1 in target luminance resulted in worse T2 performance in the later lags (i.e., after lag 2) compared to the repeat-T1 in distractor luminance in Experiment 1b. This is akin to the high-contrast condition in Chua’s, where there is more information available for acquisition. Di Lollo, Kawahara, Ghorashi & Enns (2005) recently proposed a theory that also postulated the loss of attentional control as a factor for causing the AB effect. In their study, AB was absent when the targets consisted of a stream of three items 89 belonging to the same category, but was reinstated when an intervening item of a different category was inserted between two targets. Di Lollo et al. accounted for this effect as a temporary loss of control over the prevailing attentional set, which rendered participants vulnerable to an exogenously-triggered switch in attentional set. The finding that target was a stream of three items belonging to the same category elicited no AB effect was very similar to the findings in Experiment 1a, where T2 at lag 2 was identified when the repeat-T1 at lag 1 was identical to T1. In both cases, one can argue the attentional episode that was initiated with T1 did not terminate until the third item. In other words, attention continued to be engaged as the three items because they were deemed (by the visual system) as temporally coextensive. The finding that an intervening target of a different category inserted between two targets produced an AB was similar to the findings of the repeat-T1 condition in Experiment 1b, where inserting the repeat-T1 in distractor luminance at lag 1 did not improve T2 recovery. Although Raymond (2003) interpreted her results in terms of an object file hypothesis (Kahneman, Treisman & Gibbs, 1992), her data can also be accounted within the attentional engagement hypothesis. In her study, an RSVP stream consisting of changing perspective views of the same basic object was presented. Participants had to identify the orientation of an extra feature (i.e. horizontal or vertical line) added to each target. No AB was found when T1 and T2 was the same object, but an AB was apparent when T1 and T2 were different objects. As all distractors in the RSVP stream were different perspectives of one object, it may be argued the sequential items are perceived as a single item and coded as temporally 90 coextensive. Therefore, both T1 and T2 are processed in a single object file, which eliminated the AB effect. Kellie and Shapiro (2004) also reported a similar finding where visual presentation of items in the RSVP stream were morphed smoothly or truncated.48 No AB effect was found in the former condition, while an AB effect was apparent in the latter condition. Although Kellie and Shapiro also framed their findings under an object file hypothesis (Kahneman et al., 1992), their findings can also be accounted under the attentional engagement hypothesis. For an object to maintain its integrity, the correlation obtained for sequential time slices from the sensory response function of the time-slices containing the object ought to be highly correlated. In this case, the same “object file” (Kahneman et al., 1992) is maintained. But when the correlation is dramatically lowered, the visual system interprets this as the appearance of a new object (i.e., previous object has terminated). Thus, the previous object file is closed and a new one is opened for the new object. In the context of the attentional engagement hypothesis, an “attentional episode” may be defined as attentional engagement and its subsequent disengagement (e.g., Sperling and Weichselgartner , 1995). The “object file” could be considered to be analogous to the “attentional episode”. In the smoothly morphed condition, the correlation between the sensory response functions of sequential time-slices is likely to be high. Hence, attention continues to dwell after it has been deployed to T1, hence incorporating T2 within the same attentional episode. Conversely, correlations 48 The smooth morph of items is such that there are slight changes between each items in each frame, when the initial item changes to a different item in the last frame (e.g., a smoking pipe to a saucepan). This is akin to the commonly seen computer graphic effects where a face is morphed smoothly into another face. For the truncated item stream, items did not morph smoothly as the changes from each item to item is random. However, the set of items used in both conditions are similar. 91 between sequential items in the truncated condition are likely to be lower. Hence, when attention disengages from T1, it suffers a temporally loss of attentional control (Di Lollo et al., 2005), which results in the AB effect. Attentional Control and Extant Models According to Wee and Chua (2004), there are “three possible ways in which attentional control and stimulus processing may be related” (p. 599): (a) both are simultaneous, such that attentional shift and stimulus processing begins and terminates together. This means attention continues to be engaged on a target as long as the target undergoes processing. Processing models (e.g., Chun & Potter, 1995) are an example of such a relationship. Using a spotlight metaphor, this means the spotlight (attention) is turned on whenever processing is required. It remains on until processing is completed, after which it can then be switched off and be shifted to a new target; (b) attention dwells for some unspecified period before disengaging, regardless of processing requirements. The attentional dwell hypothesis (Duncan et al., 1994) is an example. The spotlight metaphor would argue that the spotlight is switched on when processing is required. However, the amount of time the spotlight remains on is predetermined. Even if stimulus processing is completed, the spotlight cannot be switched off and be shifted to a new target until the predetermined time has lapsed; (c) attention shift and stimulus processing are controlled by different mechanisms, such that both are independent of each other. The attentional engagement hypothesis (Chua, 2005; Wee & Chua, 2004) is an example. This means 92 while the spotlight is switched on to enhance stimulus processing, it need not necessarily wait until stimulus processing is completed before it can be switched off and be shifted to a new target. Under the above conceptualization, the processing model argues that T1 processing difficulty modulates the passing of attentional control to T2. The manipulation of T1 masking effect is a manipulation of T1 processing difficulty. In most AB experiments, T1 processing difficulty is indicated by either T1 identification accuracy or T1 identification RT. In this set of studies, mean T1 identification accuracy was not correlated with AB magnitude. Also, masking does not modulate the AB in this study. This is evident in Experiment 2b, where the lowered masking effect of the chimeral distractor does not lead to a corresponding improved T2 performance. Furthermore, the four-dot distractor in Experiment 2c lowered T2 performance at lag 2 (compared to the blank condition) even though it was not supposed to mask T1. These results suggest that processing difficulty does not modulate attentional disengagement from T1. The attentional dwell model proposed by Duncan and his associates (Duncan et al., 1994; Ward et al., 1996) is conceptually similar to the attentional engagement hypothesis. In a way, the attention dwell model proposed that the AB effect is due to a failure to disengage from a previous target, as attention continues to ‘dwell’ at a target location. The attentional engagement hypothesis is a modification on the attentional dwell model, with the added assumption that attention dwell duration can be reduced when target termination is signaled to the visual system. 93 The original conceptualization of Duncan and his associates (Duncan et al., 1994; Ward et al., 1996) is not tenable as Moore et al. (1996) demonstrated that attentional dwell time is partly modulated by the masking of targets. They replicated the studies of Duncan and his associates (Duncan et al., 1994; Ward et al., 1996) but found that estimates of dwell time decreased when T1 was unmasked. In terms of the attentional engagement hypothesis, disengagement occurred earlier in the unmasked condition (i.e., when T1 was followed by a blank) because the correlation was lower. This allows attention to shift towards T2 much faster, thus reducing the amount of dwell time. Thus, contrary to Duncan et al. who claimed that attentional dwell was approximately 500 ms, Moore et al.’s finding may be reinterpreted from the attentional engagement hypothesis. Early Selection Versus Late Selection An issue central to attentional studies is whether attention is an early or late selection process. Driver and Tipper (1989) pointed out that the early selection position states that information is selected early by attention for further cognitive processing, mainly on the basis of physical characteristics (e.g., colour, shape, orientation), such that only rudimentary features are represented before attentional selection. On the other hand, the late selection position asserts that full perceptual analysis takes place for all stimuli and the selection operates subsequent to object recognition. Essentially, the major contention between both camps is whether stimuli are fully analyzed at the perception stage and then selected for task (i.e., late 94 selection), or whether stimuli are first selected for task then fully analyzed (i.e., early selection) (e.g., Allport, 1989; Pashler, 1998; Johnston & Dark, 1986). The interference model (Shapiro et al., 1994) and the two-stage processing model (Chun & Potter, 1995) explicitly state that the AB is a late selection process. The masking account of the processing model proposed by Di Lollo and associates (e.g., Seiffert & Di Lollo, 1997; Giesbrecht & Di Lollo, 1998) places the AB as an early selection process. For the attentional engagement hypothesis, it is assumed that identity of targets is unknown prior to selection. This is because as much information as possible is required to be acquired from the sensory response function before target identity can be established. Furthermore, if target identities were available prior attentional selection, the visual system would be able to detect an object change quite easily. Hence, the attentional engagement supports an early selection process. A Lower Boundary of Temporal Limits For Visual Attention? The findings in this thesis suggests that a lower boundary of temporal limit for the attending of rapidly varying stimuli, such that once engaged, attention is still unable to disengage immediately. In Experiment 2b, it was found that inserting a blank after T1 did not improved T2 performance at lag 1. One possibility was that attention was still unable to disengage from T1 at that point. Although the repeat-T1 and four-dot distractors (i.e., Experiments 2a and 2c) signal T1 termination much better to the visual system, T2 performance for these conditions were lower than the blank condition at lag 2. The difference between the repeat-T1 and four-dot 95 conditions with the blank conditions is that the T1 termination for the latter is more likely to be signaled earlier by 100 ms. Recall the attentional engagement hypothesis states that it is the correlation between the sensory response function of sequential time-slices that signals target termination. Consider the situation when the target appears in frame X. For the correlation of the sensory response function of the timeslice of frame X and that of frame X+1 to be calculated, it is necessary for the offset of frame X+1 to offset, or at least be onscreen for some time, before the above mentioned correlation could be computed. Furthermore, Dixon and Di Lollo (1994) pointed out that time is required for the calculation of this correlation. Together with the findings in this thesis, this explication of the attentional engagement hypothesis suggests a lower boundary of temporal limit for attention. The findings in Experiment 2 suggest that this lower boundary of temporal limit for attention is approximately 100 ms. Under this conceptualization, the attentional dwell model (e.g., Duncan et al., 1994; Ward et al., 1996, 1997) might be correct when it postulates that attention dwells for an unspecified amount of time regardless of processing demands. However, Duncan and his associates had greatly overestimated this lower boundary of temporal limit (i.e., they estimated to it be approximately 500 ms). Moore et al. (1996) proposed a lower estimate (i.e., approximately 150 ms), which is more or less consistent with the present findings. 96 Conclusion The central hypothesis in the current study is that the AB effect is not due to processing limitations (e.g., Chun & Potter, 1995; Seiffert & Di Lollo, 1997; Jolicœur, 1998), or interference in retrieval from VSTM (e.g. Raymond et al., 1995; Shapiro et al., 1994). Rather, the AB effect is due to a failure of attentional disengagement from T1 when T1 termination fails to be signaled to the visual system. T1 processing difficulty does not govern attentional disengagement from it. Rather, attentional disengagement from T1 is modulated by the correlation of the sensory response functions of sequential time-slices, as described by the temporal coding hypothesis (Dixon & Di Lollo, 1994). An implication here is that attention should not be viewed as a “resource” that can be exhausted by processing loads, as described by the processing models. Rather, attention should be viewed as a signal enhancing mechanism that facilitates stimulus processing. Wee and Chua (2004) argued for the need to separate between attention control and stimulus processing, which they claimed are confounded in RSVP experiments. Many previous attentional studies have failed to recognize this difference, resulting in the confounding of these two independent conceptual ideas. Another interesting theoretical issue is the conceptual definition of an object. Under the attentional engagement hypothesis, an object can be defined as visual stimuli that are highly correlated in time. Thus, the attentional engagement hypothesis could also be cast in an “object file” perspective, where AB is due to the inability of the visual system to open a second object file when a first has already been opened. In 97 other words, the attentional engagement hypothesis might be consistent with the object file metaphor proposed by Raymond (2003), and Kellie and Shapiro (2004). Although the findings in the current study can be interpreted under the attentional engagement hypothesis, the purported compatibility between the findings in the current studies and Di Lollo et al.’s (2005), Raymond’s (2003), and Kellie and Shapiro’s (2004) findings remains speculative at best. Direct testing of this compatibility is required, which is a good starting point for future researches. This will be important for the understanding of how the attentional system works. Another issue that remains to be explored is the purported lower boundary of temporal limit (a) whether this lower limit exist; and (b) if it exist, then what is a good estimate for it. Although the object file metaphor and the attentional engagement hypothesis might be consistent with each other at the moment, further research might be needed to separate both theories. Whether or not the attentional engagement hypothesis is the best theory to account for the AB effect, where the implication is that the AB effect is due to the failure to transfer attentional control to a new target, will have to await further research. 98 Bibliography Allport, D. A. (1989). Visual attention. In Posner, M. I. (Ed.), Foundations of Cognitive Science, Cambridge, Mass.: MIT Press. Arnell, K. M., & Jolicœur, P. (1999). The attentional blink across stimulus modalities: evidence for central processing limitations. Journal of Experimental Psychology: Human Perception and Performance, 25(3), 630-648. Bjork, E. L., & Murray, J. T. (1977). On the nature of input channels in visual processing. Psychological Review, 84(5), 472-484. Brehaut, J. C., Enns, J. T., & Di Lollo, V. (1999). Visual masking plays two roles in the attentional blink. Perception & Psychophysics, 61(7), 1436-1448. Breitmeyer, B. G. (1984). Visual masking: An integrative approach. New York: Oxford University Press. Breitmeyer, B. G., Ehrenstein, A., Pritchard, K., Hiscock, M., & Crisan, J. (1999). The role of location specificity and masking mechanism in the attentional blink. Perception & Psychophysics, 61(5), 798-809. 99 Broadbent, D. E., & Broadbent, M. H. P. (1987). From detection to identification: response to multiple targets in rapid serial visual presentation. Perception & Psychophysics, 42(2), 105-113. Bundesen, C. (1990). A theory of visual attention, Psychological Review, 97(4), 523547. Busey, T. A., & Loftus, G. R. (1994). Sensory and cognitive components of visual information acquisition. Psychological Review, 101(3), 446-469. Chua, F. K. (2005). The effect of target contrast on the attentional blink. Perception & Psychophysics, 67(5), 770-788. Chua, F. K., Goh, J., & Hon, N. (2000). Nature of codes extracted during the attentional blink. Journal of Experimental Psychology: Human Perception and Performance, 27(5), 1229-1242. Chun, M. M. (1997). Types and tokens in visual processing: A double dissociation between the attentional blink and repetition blindness. Journal of Experimental Psychology: Human Perception and Performance, 23(3), 738-755. 100 Chun, M. M., & Potter, M. C. (1995). A two-stage model for multiple target detection in rapid serial visual presentation. Journal of Experimental Psychology: Human Perception and Performance, 21(1), 109-127. Coltheart, M. (1980). Iconic memory and visible persistence. Perception and Psychophysics, 27, 183–228. Crebolder, J. M., Jolicœur, P., & McIlwaine, J. D. (2002). Loci of signal probability effects and of the attentional blink bottleneck. Journal of Experimental Psychology: Human Perception and Performance, 28(3), 695-716. Dell’Acqua, R., Pascali, A., Jolicœur, P.,& Sessa, P. (2003). Four-dot masking produces the attentional blink. Vision Research, 43, 1907-1913. Di Lollo, V., Kawahara, J., Ghorashi, S. M. S., & Enns, J. T. (2005). The attentional blink: Resource depletion or temporary loss of control? Psychological Research, 69, 191-200. Dixon, P., & Di Lollo, V. (1994). Beyond visible persistence: an alternative account of temporal integration and segregation in visual processing. Cognitive Psychology, 26, 33-63. 101 Driver, J. & Tipper, S. P. (1989). On the nonselectivity of selective seeing: Contrasts between interference and priming in selective attention, Journal of Experimental Psychology: Human Perception and Performance, 15(2), 304-314. Duncan, J., & Humphreys, G. W. (1989). Visual search and stimulus similarity. Psychological Review, 96(3), 443-458. Duncan, J., Ward, R., & Shapiro, K. (1994). Direct measurements of attentional dwell time in human vision. Nature, 369, 313-315. Enns, J. T., & Di Lollo, V. (1997). Object substitution: a new form of masking in unattended visual locations. Psychological Science, 8(2), 135-139. Giesbrecht, B., Bischof, W. F., & Kingstone, A. (2003). Visual masking during attentional blink: tests of the object substitution hypothesis. Journal of Experimental Psychology: Human Perception and Performance, 29(1), 238-258. Giesbrecht, B., & Di Lollo, V. (1998). Beyond the attentional blink: visual masking by object substitution. Journal of Experimental Psychology: Human Perception and Performance, 24(5), 1454-1466. 102 Grandison, T. D., Ghirerdelli, T. G., & Egeth, H. E. (1997). Beyond similarity: masking of the target is sufficient to cause the attentional blink. Perception & Psychophysics, 59(2), 266-274. Hoffman, J. E. (1978). Search through a sequentially presented visual display. Perception & Psychophysics, 23, 1-11. Isaak, M. I., Shapiro, K. L., & Martin, J. (1999). The attentional blink reflects retrieval competition among multiple rapid serial visual presentation items: tests of an interference model. Journal of Experimental Psychology: Human Perception and Performance, 25(6), 1774-1792. Jolicœur, P. (1998). Modulation of the attentional blink by on-line response selection: evidence from speeded and unspeeded task1 decisions. Memory & Cognition, 26(5), 1014-1032. Jolicœur, P. (1999a). Concurrent response-selection demands modulate the attentional blink. Journal of Experimental Psychology: Human Perception and Performance, 25(4), 1097-1113. Jolicœur, P. (1999b). Restricted attentional capacity between sensory modalities. Psychonomic Bulletin & Review, 6(1), 87-92. 103 Jolicœur, P., & Dell’Acqua, R. (2000). Selective influence of second target exposure duration and task1 load effects in the attentional blink phenomenon. Psychonomic Bulletin & Review, 7(3), 472-479. Johnston, W. A., & Dark, V. J. (1986). Selective attention, Annual Review of Psychology, 37, 43-75. Kahneman, D., Treisman, A., & Gibbs, B. J. (1992). The reviewing of object files: Object-specific integration of information. Cognitive Psychology, 24, 175-219. Kanwisher, N. (1987). Repetition Blindness: Type recognition without token individuation. Cognition, 27, 117-143. Kastner, S., & Ungerleider, L. G. (2000). Mechanism of visual attention in the human cortex. Annual Review of Neuroscience, 23, 315-341. Kawahara, J., Zuvic, S. M., Enns, J. T., & Di Lollo, V. (2003). Task switching mediates the attentional blink even without backward masking. Perception & Psychophysics, 65(3), 339-351. Kellie, F. J., & Shapiro, K. L. (2003). Object file continuity predicts attentional blink magnitude. Perception & Psychophysics, 66(4), 692-712. 104 Kristjansson, A., & Nakayama, K. (2002). The attentional blink in space and time. Vision Research, 42, 2039-2050. Lawrence, D. H. (1971). Two studies of visual search for word targets with controlled rates of presentation. Perception & Psychophysics, 10, 85-89. Loftus, G. R., Duncan, J., & Gehrig, P. (1992). On the time course of perceptual information that results from a brief visual presentation. Journal of Experimental Psychology: Human Perception and Performance, 18(2), 530-549. Loftus, G. R., & Ruthruff, E. (1994). A theory of visual information acquisition and visual memory special application to intensity-duration trade-offs. Journal of Experimental Psychology: Human Perception and Performance, 20(1), 33-49. Loftus, G. R., & Irwin, D. E. (1998). On the relations among different measures of visible and informational persistence. Cognitive Psychology, 35, 135-199. Maki, W. S., Couture, T., Frigen, L., & Lien, D. (1997). Sources of the attentional blink during rapid serial visual presentation: Perceptual interference and retrieval competition. Journal of Experimental Psychology: Human Perception & Performance, 23, 1393-1411. 105 McLaughlin, E. N., Shore, D. I., & Klein, R. M. (2001). The attentional blink is immue to masking-induced data limits. The Quarterly Journal of Experimental Psychology, 54A(1), 169-196. Moore, C. M., Egeth, H., Berglan, L. R., & Luck, S. J. (1996). Are attentional dwell times inconsistent with serial visual search? Psychonomic Bulletin & Review, 3(33), 360-365. Mozer, M. C. (1989). Types and tokens in visual letter perception. Journal or Experimental Psychology: Human Perception and Performance, 15(2), 287-303. Nakayama, K., & Mackeben, M. (1989). Sustained and transient components of focal visual attention. Vision Research, 29, 1631-1647. Neisser, U. (1967). Cognitive Psychology. New York: Appleton-Century-Crofts. Park, J., & Kanwisher, N. (1994). Determinants of repetition blindness. Journal of Experimental Psychology: Human Perception and Performance, 20(3), 500-519. Pashler, H. E. (1998). The Psychology of Attention. Cambridge, Massachusetts: MIT Press. Posner, M. I. (1980). Orienting of attention. Quarterly Journal of Experimental Psychology, 32, 3-25. 106 Posner, M. I., & Peterson, S. E. (1990). The attention system of the human brain. Annual Review of Neuroscience, 13, 25-42. Raymond, J. E. (2003). New objects, not new features, trigger the attentional blink. Psychological Science, 14, 54-59. Raymond, J. E., Shapiro, K. L., & Arnell, K. M. (1992). Temporary suppression of visual processing in an RSVP task: an attentional blink? Journal of Experimental Psychology: Human Perception and Performance, 21(3), 653-662. Raymond, J. E., Shapiro, K. L., & Arnell, K. M. (1995). Similarity determines the attentional blink. Journal of Experimental Psychology: Human Perception and Performance, 18(3), 849-860. Shiffrin, R. M., & Gardner, G. T. (1972). Visual processing capacity and attentional control. Journal of Experimental Psychology, 12, 97-136. Seiffert, A. E., & Di Lollo, V. (1997). Low-level masking in the attentional blink. Journal of Experimental Psychology: Human Perception and Performance, 23(4), 1061-1073. 107 Shapiro, K. L. (2001). Temporal methods for studying attention: how did we get here and where are we going? In K. L. Shapiro (Ed.), The limits of attention: Temporal constraints in human information processing (pp. 1-19). New York: Oxford University Press. Shapiro, K. L., Driver, J., Ward, R., & Sorensen, R. E. (1997). Priming from the attentional blink: a failure to extract visual tokens but not visual types. Psychological Science, 8(2), 95-100. Shapiro, K. L., Raymond, J. E., & Arnell, K. M. (1994). Attention to visual pattern information produces the attentional blink in rapid serial visual presentation. Journal of Experimental Psychology: Human Perception and Performance, 20(2), 357-371. Sperling, G. (1960). The information available in brief visual presentations. Psychological Monographs, 74, 1–29. Sperling, G., & Weichselgartner, E. (1995). Episodic theory of the dynamics of spatial attention. Psychological Review, 102(3), 503-532. Treisman, A, & Gelande, G. (1980). A feature integration theory of attention. Cognitive Psychology, 12, 97-136. 108 Theeuwes, J. (1994). Stimulus-driven capture and attentional set: selective search for color and visual abrupt onsets. Journal of Experimental Psychology: Human Perception and Performance, 20(4), 799-806. Visser, T. A. W., Zuvic, S. M., Bischof, W. F., & Di Lollo, V. (1999). The attentional blink with targets in different spatial locations, Psychonomic Bulletin & Review, 6(3), 432-436. Ward, R., & Duncan, J. (1996). The slow time-course of visual attention. Cognitive Psychology, 30, 79-109. Ward, R., Duncan, J., & Shapiro, K. (1997). Effects of similarity, difficulty, and nontarget presentation on the time course of visual attention. Perception & Psychophysics, 59(4), 593-600. Wee, S., & Chua, F. K. (2004). Capturing attention when attention “blinks”. Journal of Experimental Psychology: Human Perception and Performance, 30(3), 598-612. Weichselgartner, E., & Sperling, G. (1987). Dynamics of automatic and controlled visual attention. Science, 238, 778-780. 109 Wolfe, J. M., Cave, K. R., & Franzel, S. L. (1989). Guided search: An alternative to the feature integration model for visual search. Journal of Experimental Psychology: Human Perception and Performance, 15, 419-433. Yantis, S., & Hillstrom, A. P. (1994). Stimulus-driven attentional capture: evidence from equiluminant visual objects. Journal of Experimental Psychology: Human Perception and Performance, 20(1), 95-107. Yeshurun, Y., & Carrasco, M. (1998). Attention improves or impairs visual performance by enhancing visual resolution. Nature, 396, 72-74. Yeshurun, Y., & Carrasco, M. (1999). Spatial attention improves performance in spatial resolution tasks. Vision Research, 39, 293-306. Zuvic, S. M., Visser, T. A. W., & Di Lollo, V. (1999). Direct estimates of processing delays in the attentional blink. Psychological Research, 63, 192-198. 110 [...]... had the largest modulating effect on the blink, suggesting the locus of the underlying cause for the AB lies at lag 1; and (b) a blank inserted in lag 1 causes the greatest attenuation of the AB, suggesting that understanding the effects of the blank on the blink is crucial to an understanding of the underlying cause of the temporal limits of attention In this thesis, I argue that the failure to transfer... both the blank and baseline conditions are predicted for each of the AB models For the baseline condition, the lag 1 distractor is a randomly chosen letter, while the lag 1 distractor for the blank condition is a blank with a luminance similar to the background The baseline and blank conditions in the current experiment is highly similar to a typical two target RSVP presentation in the AB literature, and... argued that as a blank is highly dissimilar to both T1 and T2 (i.e., the blank contains no features), it will be assigned small or no weight Thus, the blank would not compete with the other items for attentional resources, resulting in the attenuation of the AB One might even argue that the blank would not enter into VSTM as an “item” Thus, it cannot interfere with T2 retrieval from VSTM, and this allows... effectively to the visual system In Experiment 2, three different types of lag 1 distractors are employed to test this hypothesis The magnitude of T1 termination signal is manipulated by varying the similarity between T1 and the lag 1 distractor The results from Experiment 2 support the attentional engagement hypothesis In all these experiments, the lag 1 distractor was systematically manipulated All... larger AB than baseline in Experiment 1a, and the same AB with the baseline condition in Experiment 1b Essentially, the attentional dwell model makes the same prediction as the interference model This is because both models are based on Duncan and 25 Humphrey’s (1989) theory of visual selection However, the difference between the interference and attentional dwell models is that the former is an offline... transient Unless they receive further processing and consolidation, they are subjected to rapid degradation and forgetting Items possessing target attributes (e.g colour, letter case, semantic category) are flagged They then undergo further processing in a second stage, where they are consolidated in VSTM Otherwise, their sensory representation will degrade and their identities unrecoverable Chun and... sections The situation whereby the repeat-T1 might not be treated as a distinct object due to an RB effect will also be considered in the Results and Discussion section, where I will make the claim that none of the extant AB models can account for the data of Experiments 1a and 1b, regardless of whether the repeat-T1 is treated as a distinct object from T1 or not Repeat-T1 and the Interference Model As the. .. The target is cued at the post-mask display after the presentation of the letters, with an arrow pointing to one of the columns where a letter appeared Bjork and Murray argue this serves two important functions: (a) the noiseletter interference is concentrated at a perceptual rather than a decisional level; and (b) it allows the physically identical letter to be treated as noise rather than redundant... (1997) provided further support for the attentional dwell hypothesis by replicating their basic findings using a RSVP “stream” paradigm19 According to the attentional dwell model, a blank inserted in lag 1 attenuates the AB effect because there is one fewer item to compete for the allocation of visual processing resources Therefore, the competition that results in the sustained attentional state resolves... I and O excluded Each letter appeared as a target with approximately equal frequency For each trial, the two target letters were predetermined, and then removed from the letter set The order of the remaining 22 letters was randomly shuffled The position of the first target (T1) was randomly chosen from frames 6-14 The second target (T2) was then inserted in the frame specified by the lag condition The